chunk_uid
stringlengths
40
40
chunk_type
stringclasses
2 values
chunk_index
int64
0
6.71k
total_chunks
int64
1
6.71k
section_title
stringlengths
1
157
embed_text
stringlengths
1
83.3k
spans
dict
paper_doi
stringlengths
0
63
paper_id_arxiv
stringlengths
9
16
title
stringlengths
7
245
authors
listlengths
1
768
categories
listlengths
1
7
year
int64
2k
2.02k
language
stringclasses
2 values
discipline
stringclasses
8 values
sparse_indices
listlengths
1
1.02k
sparse_values
listlengths
1
1.02k
3f50ea0bce3946f4a22b2df4ebf49bcc88bc9961
subsection
45
83
Wired spanning forest oriented toward a root
Then, for \pi -almost every initial rotor configuration, the scaled walk (\frac{1}{\sqrt{n}} X_{\lfloor nt\rfloor } )_{t \ge 0} converges weakly on D_{\mathbb {R}^d}[0,\infty ) to a Brownian motion with diffusion matrix \Gamma . \end{} }}As a corollary of Theorem~\ref {scaling limit for elliptic rotor walk}, we derive ...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 100, 1434, 840, 11907, 61475, 2062, 1290, 180346, 105994, 71, 35691, 132076, 418, 864, 3198, 19, 1193, 17463, 365, 52347, 429, 757, 158, 74789, 642, 344, 98, 391, 454, 1052, 2389, 46632, 47, 40218, 3378, 78112, 45755, 92105, 50944, 36557,...
[ 0.017791748046875, 0.2008056640625, 0.088623046875, 0.1273193359375, 0.14111328125, 0.1275634765625, 0.16455078125, 0.209228515625, 0.242431640625, 0.1240234375, 0.260498046875, 0.128662109375, 0.04754638671875, 0.1317138671875, 0.162841796875, 0.08477783203125, 0.1112060546875, 0....
99ca79bede64bbc082d3cb11b59d53402f0c1b23
subsection
46
83
Wired spanning forest oriented toward a root
A stationary distribution \pi of M is \emph {ergodic} if {\pi }[{B}]\in \lbrace 0,1\rbrace for any invariant set {B}. }}}}}}}}Let \Omega ^{\mathbb {N}} be the \emph {trajectory space} of M, \Omega ^{\mathbb {N}}:=\lbrace (\omega _i)_{i \ge 0} \mid \omega _i \in \Omega \rbrace , equipped with the product \sigma -algebr...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 62, 29398, 6635, 113068, 41872, 1434, 111, 276, 83, 6, 195, 11727, 56, 519, 18403, 8152, 2174, 10666, 51912, 1065, 24854, 571, 268, 73, 99407, 72966, 100, 2499, 23, 162591, 5423, 124480, 670, 87849, 5125, 839, 47391, 186, 70, 75161, 326...
[ 0.06207275390625, 0.25244140625, 0.1644287109375, 0.295654296875, 0.01385498046875, 0.273681640625, 0.07421875, 0.1802978515625, 0.1142578125, 0.004486083984375, 0.01837158203125, 0.181884765625, 0.07635498046875, 0.1744384765625, 0.256591796875, 0.004974365234375, 0.0703125, 0.004...
a0bd60354b6ff7e89c819601e7966ca0c4559512
subsection
47
83
Wired spanning forest oriented toward a root
Now note that, for any n \ge 1, \begin{align*} \pi [{B}] &= \int _{\Omega } P^{(n)}(x, {B}) \ d\pi (x) \qquad \text{(by the stationarity of $\pi $)}\\ &=\int _{{B}} P^{(n)}(x,{B}) \ d\pi (x) + \int _{{B}^{\prime } \setminus {B}} P^{(n)}(x, {B}) \ d\pi (x) \qquad \text{(as $P^{(n)}(x, {B}) = 0$ for $x \notin {B}^{\prime...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 20537, 2499, 653, 6, 429, 106, 372, 6820, 143420, 8152, 1434, 24854, 571, 268, 619, 1369, 41872, 4288, 87849, 51912, 436, 8353, 132, 19, 16, 425, 4, 104, 15, 91526, 22829, 70, 29398, 147, 2481, 111, 3650, 13273, 101, 172162, 47391, 99...
[ 0.031524658203125, 0.085205078125, 0.068115234375, 0.010650634765625, 0.1318359375, 0.05218505859375, 0.0106201171875, 0.0064697265625, 0.006195068359375, 0.010955810546875, 0.25048828125, 0.010650634765625, 0.10986328125, 0.010528564453125, 0.018524169921875, 0.01055908203125, 0.010...
0a3c977980c0dcca3b194a32c9db7a4841da16f8
subsection
48
83
Wired spanning forest oriented toward a root
We write {{{C}}:= \lbrace \rho \in \text{Rot}(G) \mid \exists ~\rho ^{\prime } \in {{{B}}\text{ s.t. } \rho \text{ and } \rho ^{\prime } \text{ differ at finitely many vertices}\rbrace . Note that {{{C}} is a tail event that contains {{{B}}. It then suffices to show that \pi [{{{C}}\setminus {{{B}}]=0. }}Let \rho be an...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 1401, 33022, 76339, 441, 47391, 12, 1369, 41872, 99407, 6, 42, 497, 22829, 12724, 18, 8152, 724, 16, 22000, 54376, 7, 6780, 13331, 24854, 114654, 51912, 571, 91, 5, 136, 129927, 35108, 37838, 5941, 95378, 5170, 83, 10, 46741, 19732, 450...
[ 0.0755615234375, 0.19970703125, 0.03997802734375, 0.1270751953125, 0.011474609375, 0.030609130859375, 0.0478515625, 0.0113525390625, 0.16259765625, 0.012542724609375, 0.0164947509765625, 0.228759765625, 0.1083984375, 0.120849609375, 0.0870361328125, 0.0113525390625, 0.156494140625, ...
6086505b09bf677484e0b1cf549213c6aea92127
subsection
49
83
Wired spanning forest oriented toward a root
(d) The final rotor configuration of the RWLM at the end of this process, which is the same regardless of whether the initial configuration is (a) or (b). } \end{} }}}Write ”:=xn-1(n)=xn-1(n'). Note that ” is the rotor configuration at the n-th step of the scenery process if the walker of the RWLM follows x0,...,xn a...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 71, 2704, 2062, 1290, 180346, 70, 78691, 37150, 99, 3564, 903, 9433, 4, 83, 5701, 211460, 111, 36766, 61475, 15, 11, 16, 707, 275, 51912, 6, 41872, 24854, 8152, 18781, 425, 19, 5759, 1369, 132, 18622, 970, 653, 9, 927, 29954, 28302, ...
[ 0.0950927734375, 0.18701171875, 0.1693115234375, 0.19921875, 0.256591796875, 0.035400390625, 0.161376953125, 0.2408447265625, 0.05804443359375, 0.181884765625, 0.031158447265625, 0.18505859375, 0.011322021484375, 0.039215087890625, 0.12890625, 0.0931396484375, 0.011566162109375, 0....
adc307f02842c801e042be5bd677c0be8c296536
subsection
50
83
Wired spanning forest oriented toward a root
Recall that X_n denotes the walker^{\prime }s location and \rho _n denotes the rotor configuration at the n-th step of the RWLM. }}}\begin{} Let (G,c) be a weighted lattice graph in \mathbb {R}^d, and suppose that the mechanism of the RWLM satisfies \ref {item: TR} and \ref {item: ELL}. Let \pi be any tail trivial stat...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 85763, 1193, 454, 19, 8, 157, 35691, 56, 114654, 31913, 497, 101, 2062, 1290, 180346, 653, 927, 29954, 78691, 37150, 10842, 724, 238, 186, 57888, 10495, 24494, 41382, 71, 139124, 191619, 40407, 3387, 29087, 217, 14107, 56051, 1434, 2499, ...
[ 0.102294921875, 0.10589599609375, 0.11279296875, 0.1656494140625, 0.0906982421875, 0.1177978515625, 0.1651611328125, 0.16748046875, 0.1165771484375, 0.1439208984375, 0.186767578125, 0.007904052734375, 0.1005859375, 0.1292724609375, 0.1519775390625, 0.0714111328125, 0.0428466796875, ...
bc2dc532fba7094c2ca231bcc3073178c5ef89a0
subsection
51
83
Wired spanning forest oriented toward a root
It follows from Definition REF and that V_i=& X_{i+1}- X_{i} = \sum _{\mathbf {y}\in {N}(\operatorname{\mathbf {0}})} \mathbb {1}\lbrace \rho _{i}(X_i)-X_i=\mathbf {y}\rbrace \, Y_{\mathbf {y},i}, where Y_{\mathbf {y},i} is a random variable on neighbors of the origin sampled from p_{\operatorname{\mathbf {0}}}(\math...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 1650, 28960, 7, 1295, 155455, 9069, 919, 136, 450, 310, 454, 14, 1369, 1230, 1193, 21748, 9, 2203, 41872, 11832, 150598, 53, 73, 839, 206469, 11627, 125458, 2389, 5125, 418, 99407, 42, 497, 1542, 990, 83, 96759, 77336, 98, 208244, 59665...
[ 0.1634521484375, 0.140380859375, 0.02410888671875, 0.050537109375, 0.201171875, 0.213134765625, 0.290771484375, 0.03179931640625, 0.06048583984375, 0.284912109375, 0.165283203125, 0.2376708984375, 0.1014404296875, 0.1058349609375, 0.1749267578125, 0.209228515625, 0.1358642578125, 0...
6b58ffb0ac8781de15ced4e8a6700500f6458415
subsection
52
83
Wired spanning forest oriented toward a root
Since \mu _{\operatorname{\mathbf {0}}} is a stationary distribution of the mechanism at \operatorname{\mathbf {0}} by , it then follows that: \lim _{n \rightarrow \infty } \frac{1}{n}\sum _{i=0}^{n-1} \mathbb {E}\mathopen {}\mathclose {\left[V_i V_i^\top \mid {F}_i\right]} =&\sum _{\mathbf {y}^{\prime } \in {N}(\oper...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1002/rsa.20747", "end": 1940, "openalex_id": "https://openalex.org/W2962799388", "raw": "Wilfried Huss, Lionel Levine, and Ecaterina Sava-Huss, Interpolating between random walk and rotor walk, Random structures & algorithms 52 (2018), n...
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 66016, 41872, 561, 206469, 11627, 125458, 150598, 2389, 8152, 83, 10, 29398, 6635, 113068, 191619, 99, 6, 24854, 47391, 390, 4, 28960, 7, 3141, 54969, 118201, 46632, 51912, 132076, 418, 19, 11832, 101, 14, 145407, 8353, 5759, 5125, 10666, ...
[ 0.013275146484375, 0.013885498046875, 0.2088623046875, 0.1640625, 0.1207275390625, 0.0223388671875, 0.1580810546875, 0.13525390625, 0.01348876953125, 0.061767578125, 0.014007568359375, 0.1890869140625, 0.088623046875, 0.2462158203125, 0.224853515625, 0.0579833984375, 0.01385498046875...
f10281a1d32c98bf9eed9b401171b7062702db1b
subsection
53
83
Wired spanning forest oriented toward a root
Let {{{F}}:={{{F}}(G) \subseteq 2^{{{E}(G)} be the \sigma -algebra on the set of oriented subgraphs of G generated by sets of the form \lbrace {{H}\in 2^{{{E}(G)} \mid {{{B}}\subseteq {{H}\rbrace , where {{{B}} is a finite subset of {{E}(G). The oriented wired spanning forest will be a probability distribution on the m...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 10842, 919, 47391, 724, 16, 6, 22144, 59155, 864, 116, 24854, 647, 8152, 186, 70, 20561, 192, 289, 429, 2844, 5423, 111, 23509, 297, 1614, 41382, 7, 527, 139392, 71, 390, 3173, 99407, 841, 41872, 73, 8353, 172162, 132, 22000, 76339, 5...
[ 0.075927734375, 0.094482421875, 0.013031005859375, 0.13134765625, 0.012359619140625, 0.01220703125, 0.11474609375, 0.121826171875, 0.012176513671875, 0.0693359375, 0.0125732421875, 0.09075927734375, 0.01226806640625, 0.03582763671875, 0.012176513671875, 0.150390625, 0.1580810546875, ...
1f2f5707bee49d0ab720ca5bc3e5d98e010440c8
subsection
54
83
Wired spanning forest oriented toward a root
The \emph {$r$-oriented wired spanning forest}, denoted ${{\operatorname{\mathsf {{WSF}}}}_r:={{\operatorname{\mathsf {{WSF}}}}_r(G,c), is a probability distribution on oriented subgraphs of G such that, for any wired exhaustion and any finite {{{B}}\subseteq {{E}(G), \begin{equation} {{\operatorname{\mathsf {{WSF}}}}_...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 581, 41872, 195, 11727, 42, 49133, 297, 1439, 175884, 100, 525, 157, 206469, 420, 52143, 919, 724, 238, 83, 37242, 2481, 113068, 23509, 1614, 41382, 527, 6044, 2499, 146353, 94418, 571, 5490, 3141, 757, 561, 96759, 15917, 1690, 53201, 19,...
[ 0.0263824462890625, 0.03961181640625, 0.001373291015625, 0.182373046875, 0.134033203125, 0.2301025390625, 0.0958251953125, 0.1832275390625, 0.24169921875, 0.1876220703125, 0.1966552734375, 0.0758056640625, 0.0367431640625, 0.16943359375, 0.1356201171875, 0.2193603515625, 0.1654052734...
053d14ebe64f47b734eaac9121b24c2dc48cc30b
subsection
55
83
Wired spanning forest oriented toward a root
However, it is always an \emph {r-oriented spanning forest} of G: the underlying graph of {{F} is a spanning forest, every vertex in V(G)\setminus \lbrace r\rbrace has outdegree 1 in {{F}, and r has outdegree 0 in {{F}. The first condition follows from Lemma~\ref {lemma: wsf and oriented wsf}, and the others can be ver...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 33306, 442, 83, 11343, 142, 6, 41872, 195, 11727, 42, 49133, 297, 175884, 100, 525, 111, 527, 12, 70, 1379, 538, 214, 41382, 919, 10, 11907, 493, 24371, 310, 724, 16, 3509, 34731, 7, 99407, 1690, 1556, 1810, 112, 7134, 13, 106, 23, ...
[ 0.08453369140625, 0.018829345703125, 0.054931640625, 0.1727294921875, 0.043304443359375, 0.0168914794921875, 0.037994384765625, 0.105224609375, 0.21533203125, 0.11962890625, 0.2254638671875, 0.087646484375, 0.20947265625, 0.14501953125, 0.162109375, 0.052825927734375, 0.211669921875,...
e214c6c6b428a7637e266826e6628a9393e4399a
subsection
56
83
Wired spanning forest oriented toward a root
\item Suppose that {{{T}}(i) has been generated. Start an independent network random walk at x_{i+1} and stop it at the first time it hits {{{T}}(i) (note that the random walk hits {{{T}}(i) a.s.\ by recurrence). Let \langle y_0,\ldots , y_m \rangle be the loop erasure of this random walk. \item Set {{{T}}(i+1) to be t...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 41872, 217, 195, 121691, 8364, 76339, 618, 47391, 14, 1556, 2809, 139392, 71, 13794, 41371, 33120, 96759, 35691, 99, 1022, 21748, 8152, 7279, 70, 5117, 1733, 5962, 7, 16, 10, 390, 456, 64240, 6620, 10842, 6, 3066, 133, 113, 2389, 30591,...
[ 0.04754638671875, 0.1036376953125, 0.05853271484375, 0.06549072265625, 0.03887939453125, 0.04632568359375, 0.1292724609375, 0.03857421875, 0.0855712890625, 0.074462890625, 0.0177764892578125, 0.1851806640625, 0.03106689453125, 0.09326171875, 0.15234375, 0.17138671875, 0.181884765625,...
5f5018a8c9661e8a8b25543f2903bb02877f9cb5
subsection
57
83
Wired spanning forest oriented toward a root
Then for any finite subset {{{B}} of {{E}(G), any ordering of V(G) \setminus \lbrace r\rbrace , and any wired exhaustion of G, \mathbb {P}[{{{B}}\subseteq {{{T}}] =\lim _{n \rightarrow 0} \ {{\mu _{r,n}}[{{{B}}\subseteq {{{T}}_n], where {{{T}} is a random tree of G generated using Wilson^{\prime }s method, with root ...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 47009, 100, 2499, 94418, 13, 1614, 3509, 571, 47391, 647, 724, 12989, 214, 111, 310, 16, 6, 34731, 7, 99407, 1690, 4, 1439, 297, 146353, 1830, 527, 41872, 125458, 5125, 10666, 683, 8152, 1065, 24854, 22144, 59155, 864, 76339, 618, 268, ...
[ 0.018341064453125, 0.0880126953125, 0.097900390625, 0.1903076171875, 0.051727294921875, 0.173095703125, 0.171142578125, 0.06854248046875, 0.019744873046875, 0.101806640625, 0.1292724609375, 0.19873046875, 0.03253173828125, 0.036346435546875, 0.166748046875, 0.0198211669921875, 0.0195...
ac3e93285e237ccc2d8d84be9ed5769c4885f0c0
subsection
58
83
Wired spanning forest oriented toward a root
Start a network random walk at x_{i+1}. Stop the walk the first time it hits {{{F}}(i); if it never hits {{{F}}(i) then let it run indefinitely. This walk is locally finite a.s.\ by transience. Let \langle y_0^{\prime },y_1^{\prime },\ldots \rangle be the loop erasure of this random walk. }\item Set {{{F}}(i+1) to be t...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 13794, 10, 33120, 96759, 35691, 99, 1022, 14, 21748, 5, 42284, 70, 5117, 1733, 5962, 7, 919, 132, 2174, 8306, 76339, 47391, 16, 2633, 11675, 49919, 37838, 4000, 538, 94418, 13, 390, 3900, 6620, 10842, 6, 3066, 133, 113, 454, 2389, 835...
[ 0.1427001953125, 0.012176513671875, 0.2191162109375, 0.205810546875, 0.2384033203125, 0.07135009765625, 0.0377197265625, 0.08837890625, 0.15185546875, 0.012237548828125, 0.157958984375, 0.045867919921875, 0.042236328125, 0.025848388671875, 0.1463623046875, 0.0252227783203125, 0.11193...
d1d6638d0d293f5e1aa2f8453ef09cf5d3516072
subsection
59
83
Wired spanning forest oriented toward a root
That is, if \operatorname{{\mathsf {LE}}}\langle x_i \mid i \le I\rangle =\langle y_{I,i} \mid i \le m_I\rangle and \operatorname{{\mathsf {LE}}}\langle x_i \mid i \ge 0 \rangle =\langle y_{i} \mid i \ge 0\rangle , then for every i and all sufficiently large I we have y_{I,i}=y_i. Since G is transient, it follows that ...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 9925, 83, 2174, 206469, 11627, 41872, 420, 15300, 8152, 47391, 3066, 133, 1022, 454, 14, 6, 22000, 17, 87, 5445, 113, 568, 4, 347, 136, 125458, 7, 10666, 429, 757, 2203, 24854, 11907, 756, 129980, 538, 21334, 765, 1369, 53, 66016, 527...
[ 0.001434326171875, 0.07867431640625, 0.0889892578125, 0.1907958984375, 0.1158447265625, 0.005645751953125, 0.1468505859375, 0.16943359375, 0.00732421875, 0.006011962890625, 0.15966796875, 0.1199951171875, 0.1429443359375, 0.1114501953125, 0.1463623046875, 0.045166015625, 0.1451416015...
4ad9fc1def2c63bf5b410fb032abcfe9a491c5f3
subsection
60
83
Wired spanning forest oriented toward a root
It then follows from definition of oriented wired spanning forest for finite graphs (Definition~\ref {definition: wsf finite graphs}) that {{{T}}_n has the law of {{\operatorname{\mathsf {{WSF}}}}_{r}(G_n,c_n). }Let \tau ^j_n be the first time that \langle X_i^j \mid i \ge 0\rangle reaches the portion of the spanning t...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 28960, 80934, 23509, 297, 1439, 175884, 100, 525, 94418, 13, 6, 41382, 7, 187423, 943, 1363, 29087, 49919, 148, 420, 8152, 16, 618, 47391, 454, 19, 1556, 27165, 111, 206469, 11627, 24854, 41872, 125458, 76339, 52143, 919, 42, 132, 724, ...
[ 0.0894775390625, 0.142578125, 0.240234375, 0.113037109375, 0.184814453125, 0.24072265625, 0.178466796875, 0.1866455078125, 0.1942138671875, 0.083740234375, 0.020599365234375, 0.2349853515625, 0.0374755859375, 0.02142333984375, 0.004791259765625, 0.0208892822265625, 0.1014404296875, ...
965da278e15f42e28da20e32883cabd9949508b4
subsection
61
83
Wired spanning forest oriented toward a root
Starting an RWLM from a native environment allows us to use ergodic theory in Section~\ref {SLLN}. The main result of this section is Theorem~\ref {stationarity theorem}, which gives an explicit distribution as a native environment for the RWLM. }\end{equation}In this section the underlying graph of the RWLM will be a ...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 13794, 78691, 37150, 1295, 24, 4935, 65998, 114864, 4527, 37443, 18403, 154453, 140978, 29087, 30508, 81795, 5201, 16750, 40059, 581, 58391, 41311, 147, 2481, 70, 143726, 113068, 5490, 4153, 1379, 538, 41382, 145612, 4293, 35108, 139392, 21115,...
[ 0.1104736328125, 0.1591796875, 0.2464599609375, 0.052886962890625, 0.1302490234375, 0.1324462890625, 0.18310546875, 0.036865234375, 0.0721435546875, 0.1920166015625, 0.1834716796875, 0.179931640625, 0.09625244140625, 0.1365966796875, 0.0184783935546875, 0.1383056640625, 0.04159545898...
94bd2749027dd2bda13142971cb9b290e1fe07a5
subsection
62
83
Wired spanning forest oriented toward a root
Note that the measure \mu _{\operatorname{\mathnormal {o}}} is symmetric (i.e., \mu _{\operatorname{\mathnormal {o}}}(x)=\mu _{\operatorname{\mathnormal {o}}}(x^{-1})) as a consequence of c: \mathcal {S}\rightarrow \mathbb {R}_{>0} being symmetric. }}{{}{\bfseries {Definition 5.16 (Transitive mechanism)}} Let $(G,c)$ ...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 18622, 72350, 561, 206469, 11627, 125458, 33176, 31, 83, 230612, 238, 425, 5759, 179804, 501, 294, 54969, 1052, 2389, 8035, 150598, 32832, 187423, 943, 1892, 2485, 62624, 14, 191619, 724, 57888, 297, 145612, 4293, 41382, 48543, 276, 99407, ...
[ 0.008270263671875, 0.236328125, 0.1729736328125, 0.1175537109375, 0.043914794921875, 0.055877685546875, 0.2186279296875, 0.1239013671875, 0.038818359375, 0.2369384765625, 0.09552001953125, 0.08154296875, 0.1026611328125, 0.05938720703125, 0.06982421875, 0.06982421875, 0.0215759277343...
b58e1a1eeac3c968275724ea5023901bcea63026
subsection
63
83
Wired spanning forest oriented toward a root
\end{}}The scenery process $(\widehat{\rho }_n)_{n \ge 0}$ is a Markov chain with state space the set of rotor configurations of $G$ and with transition rule \begin{equation} \widehat{\rho }_{n+1}(x) := {\mathopen {}\mathclose {\left\lbrace \begin{array}{ll}\operatorname{\mathnormal {o}}&\text{ if } x= Y_{n}^{-1};\\ Y_...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 3611, 3957, 28302, 1294, 9433, 113458, 2943, 497, 51912, 19, 16, 24854, 6, 429, 757, 8152, 4369, 83, 10, 7880, 515, 121293, 678, 11341, 32628, 5423, 111, 2062, 1290, 180346, 7, 724, 149307, 79986, 41872, 372, 6820, 13, 5490, 2320, 454, ...
[ 0.0160369873046875, 0.06829833984375, 0.281005859375, 0.2587890625, 0.27734375, 0.115234375, 0.1875, 0.172607421875, 0.0167999267578125, 0.0758056640625, 0.0168304443359375, 0.0167694091796875, 0.0167388916015625, 0.1285400390625, 0.11865234375, 0.0167694091796875, 0.0167083740234375...
6a92af2fb838ff71ddcd66d419b57d3c318e9ea4
subsection
64
83
Wired spanning forest oriented toward a root
The \emph {$r$-oriented wired spanning forest plus one edge}, denoted ${{\operatorname{\mathsf {{WSF}}}}_r^+:={{\operatorname{\mathsf {{WSF}}}}_r^+(G,c), is the law of the random subgraph {{{F}}\sqcup \lbrace (r, Y)\rbrace , where {{{F}} is a random r-oriented forest of G sampled from {{\operatorname{\mathsf {{WSF}}}}_...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 581, 41872, 195, 11727, 42, 49133, 297, 1439, 175884, 100, 525, 1001, 1632, 121303, 157, 3674, 206469, 11627, 24854, 125458, 7, 420, 52143, 919, 47391, 454, 8353, 1328, 12, 1369, 172162, 76339, 132, 724, 238, 83, 27165, 96759, 1614, 41382...
[ 0.036346435546875, 0.06390380859375, 0.040283203125, 0.2115478515625, 0.1099853515625, 0.205810546875, 0.06744384765625, 0.1607666015625, 0.2279052734375, 0.1776123046875, 0.1937255859375, 0.15673828125, 0.1156005859375, 0.2210693359375, 0.06591796875, 0.011871337890625, 0.1032714843...
f7f9b2fe517db2bd2e368dc78a532a414218b099
subsection
65
83
Wired spanning forest oriented toward a root
Each unicycle {{{U}} is picked with probability proportional to {\Xi ({{{U}})}, where \Xi ({{{U}}) is as in Definition~\ref {definition: weight of a directed tree}. This implies that {{F_{Y}} \, \sqcup \, \lbrace (Y, r) \rbrace is distributed as {{\operatorname{\mathsf {{WSF}}}}^+_r, as desired. } }}\begin{}[Proof of T...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 98423, 19844, 75457, 1062, 83, 171739, 678, 37242, 2481, 123875, 1542, 14, 237, 155455, 29087, 49919, 57888, 111, 8951, 53201, 35388, 919, 454, 1723, 864, 33874, 99407, 1690, 15917, 125458, 420, 52143, 1328, 42, 104851, 10752, 581, 58391, 4...
[ 0.146728515625, 0.169921875, 0.269287109375, 0.1400146484375, 0.067138671875, 0.2032470703125, 0.053314208984375, 0.2000732421875, 0.115966796875, 0.1671142578125, 0.077392578125, 0.07861328125, 0.0296173095703125, 0.1072998046875, 0.099609375, 0.0565185546875, 0.23193359375, 0.008...
4ada44daa7795e8b8dbadc02efe3d7d83012eee2
subsection
66
83
Wired spanning forest oriented toward a root
It now follows from Lemma~\ref {lemma: wsfplus sum} that \widehat{\rho }_1 is distributed according to {{\operatorname{\mathsf {{WSF}}}}^+_{\operatorname{\mathnormal {o}}}, and the proof is complete. } }An important property of {{\operatorname{\mathsf {{WSF}}}}_r^+ is that it is a tail trivial measure, defined below. T...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 5036, 28960, 636, 18023, 29087, 6153, 192, 148, 7, 420, 32108, 10554, 113458, 2943, 497, 115187, 15917, 59499, 206469, 125458, 52143, 919, 1328, 33176, 31, 98869, 28484, 5526, 57266, 11627, 42, 8353, 83, 46741, 1927, 686, 289, 72350, 61924,...
[ 0.0295257568359375, 0.1068115234375, 0.02813720703125, 0.1904296875, 0.159912109375, 0.08349609375, 0.1297607421875, 0.07440185546875, 0.044097900390625, 0.1591796875, 0.25146484375, 0.2073974609375, 0.172119140625, 0.2244873046875, 0.1956787109375, 0.1571044921875, 0.277099609375, ...
2f6f913a22ef70a9cab662efd2b2908aa6371f1c
subsection
67
83
Wired spanning forest oriented toward a root
Also note that {{\operatorname{\mathsf {{WSF}}}}_r[g({{{B}})]=\operatorname{\mathsf {{WSF}}}[f\circ g({{{B}})] by Lemma~\ref {lemma: wsf and oriented wsf}. Finally, note that the set f \circ g({{{B}}) is a tail event in {F} since {{{B}} is a tail event in {{{F}}. The conclusion of the proposition now follows from the t...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 20537, 76339, 206469, 11627, 125458, 7, 420, 52143, 919, 47391, 42, 177, 132, 24854, 571, 16, 268, 1369, 41872, 8152, 1065, 82063, 706, 390, 636, 18023, 2306, 29087, 10666, 6153, 192, 148, 136, 23509, 297, 4, 70, 5423, 1238, 6, 83, 46...
[ 0.0770263671875, 0.018890380859375, 0.133544921875, 0.0911865234375, 0.055938720703125, 0.014862060546875, 0.151123046875, 0.0921630859375, 0.14453125, 0.0113525390625, 0.03802490234375, 0.0902099609375, 0.011199951171875, 0.011322021484375, 0.0740966796875, 0.0113525390625, 0.011413...
9dce6df3864fd14d9641d30de830edd1400460b8
subsection
68
83
Wired spanning forest oriented toward a root
\end{} We will further assume that the initial rotor configuration of the RWLM is sampled from a {tail trivial} native environment~(Definitions~\ref {definition: native environment} and \ref {definition: tail trivial}). }}Several remarks are in order. Condition \ref {item: ELL} is known as an \emph {ellipticity} condit...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 3611, 53333, 41591, 61475, 2062, 1290, 180346, 78691, 37150, 121413, 1295, 46741, 1927, 686, 289, 24, 4935, 65998, 187423, 943, 49919, 29087, 88832, 12989, 46347, 217, 56051, 51529, 142, 195, 11727, 24134, 112569, 2481, 35431, 163136, 96759, ...
[ 0.0111083984375, 0.01519775390625, 0.10711669921875, 0.111083984375, 0.1561279296875, 0.1795654296875, 0.207275390625, 0.1839599609375, 0.25439453125, 0.1226806640625, 0.0037841796875, 0.2303466796875, 0.08935546875, 0.1539306640625, 0.10418701171875, 0.1185302734375, 0.1107177734375...
ee97da6b93ecc187e8271b39b9710bb33d8bd2b1
subsection
69
83
Wired spanning forest oriented toward a root
Then, for \pi -almost every initial rotor configuration, the scaled walk (\frac{1}{\sqrt{n}} X_{\lfloor nt\rfloor } )_{t \ge 0} converges weakly on D_{\mathbb {R}^d}[0,\infty ) to a Brownian motion with diffusion matrix \Gamma . \end{} }}As a corollary of Theorem~\ref {scaling limit for elliptic rotor walk}, we derive ...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 100, 1434, 840, 11907, 61475, 2062, 1290, 180346, 105994, 71, 35691, 132076, 418, 864, 3198, 19, 1193, 17463, 365, 52347, 429, 757, 158, 74789, 642, 344, 98, 391, 454, 1052, 2389, 46632, 47, 40218, 3378, 78112, 45755, 92105, 50944, 36557,...
[ 0.017791748046875, 0.2008056640625, 0.088623046875, 0.1273193359375, 0.14111328125, 0.1275634765625, 0.16455078125, 0.209228515625, 0.242431640625, 0.1240234375, 0.260498046875, 0.128662109375, 0.04754638671875, 0.1317138671875, 0.162841796875, 0.08477783203125, 0.1112060546875, 0....
4ff9a65dd79dd8760a5a5e89ead0961b034a5d56
subsection
70
83
Wired spanning forest oriented toward a root
A stationary distribution \pi of M is \emph {ergodic} if {\pi }[{B}]\in \lbrace 0,1\rbrace for any invariant set {B}. }}}}}}}}Let \Omega ^{\mathbb {N}} be the \emph {trajectory space} of M, \Omega ^{\mathbb {N}}:=\lbrace (\omega _i)_{i \ge 0} \mid \omega _i \in \Omega \rbrace , equipped with the product \sigma -algebr...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 62, 29398, 6635, 113068, 41872, 1434, 111, 276, 83, 6, 195, 11727, 56, 519, 18403, 8152, 2174, 10666, 51912, 1065, 24854, 571, 268, 73, 99407, 72966, 100, 2499, 23, 162591, 5423, 124480, 670, 87849, 5125, 839, 47391, 186, 70, 75161, 326...
[ 0.06207275390625, 0.25244140625, 0.1644287109375, 0.295654296875, 0.01385498046875, 0.273681640625, 0.07421875, 0.1802978515625, 0.1142578125, 0.004486083984375, 0.01837158203125, 0.181884765625, 0.07635498046875, 0.1744384765625, 0.256591796875, 0.004974365234375, 0.0703125, 0.004...
d99bddf1951b6ecb59df5ab52d46537853d3b853
subsection
71
83
Wired spanning forest oriented toward a root
Now note that, for any n \ge 1, \begin{align*} \pi [{B}] &= \int _{\Omega } P^{(n)}(x, {B}) \ d\pi (x) \qquad \text{(by the stationarity of $\pi $)}\\ &=\int _{{B}} P^{(n)}(x,{B}) \ d\pi (x) + \int _{{B}^{\prime } \setminus {B}} P^{(n)}(x, {B}) \ d\pi (x) \qquad \text{(as $P^{(n)}(x, {B}) = 0$ for $x \notin {B}^{\prime...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 20537, 2499, 653, 6, 429, 106, 372, 6820, 143420, 8152, 1434, 24854, 571, 268, 619, 1369, 41872, 4288, 87849, 51912, 436, 8353, 132, 19, 16, 425, 4, 104, 15, 91526, 22829, 70, 29398, 147, 2481, 111, 3650, 13273, 101, 172162, 47391, 99...
[ 0.031524658203125, 0.085205078125, 0.068115234375, 0.010650634765625, 0.1318359375, 0.05218505859375, 0.0106201171875, 0.0064697265625, 0.006195068359375, 0.010955810546875, 0.25048828125, 0.010650634765625, 0.10986328125, 0.010528564453125, 0.018524169921875, 0.01055908203125, 0.010...
5dd165ae15b225984e0d63d76478300ca61b9b00
subsection
72
83
Wired spanning forest oriented toward a root
We write {{{C}}:= \lbrace \rho \in \text{Rot}(G) \mid \exists ~\rho ^{\prime } \in {{{B}}\text{ s.t. } \rho \text{ and } \rho ^{\prime } \text{ differ at finitely many vertices}\rbrace . Note that {{{C}} is a tail event that contains {{{B}}. It then suffices to show that \pi [{{{C}}\setminus {{{B}}]=0. }}Let \rho be an...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 1401, 33022, 76339, 441, 47391, 12, 1369, 41872, 99407, 6, 42, 497, 22829, 12724, 18, 8152, 724, 16, 22000, 54376, 7, 6780, 13331, 24854, 114654, 51912, 571, 91, 5, 136, 129927, 35108, 37838, 5941, 95378, 5170, 83, 10, 46741, 19732, 450...
[ 0.0755615234375, 0.19970703125, 0.03997802734375, 0.1270751953125, 0.011474609375, 0.030609130859375, 0.0478515625, 0.0113525390625, 0.16259765625, 0.012542724609375, 0.0164947509765625, 0.228759765625, 0.1083984375, 0.120849609375, 0.0870361328125, 0.0113525390625, 0.156494140625, ...
ffe2a4eca5b04dd1b7da917355093c2d3d7c576a
subsection
73
83
Wired spanning forest oriented toward a root
(d) The final rotor configuration of the RWLM at the end of this process, which is the same regardless of whether the initial configuration is (a) or (b). } \end{} }}}Write ”:=xn-1(n)=xn-1(n'). Note that ” is the rotor configuration at the n-th step of the scenery process if the walker of the RWLM follows x0,...,xn a...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 71, 2704, 2062, 1290, 180346, 70, 78691, 37150, 99, 3564, 903, 9433, 4, 83, 5701, 211460, 111, 36766, 61475, 15, 11, 16, 707, 275, 51912, 6, 41872, 24854, 8152, 18781, 425, 19, 5759, 1369, 132, 18622, 970, 653, 9, 927, 29954, 28302, ...
[ 0.0950927734375, 0.18701171875, 0.1693115234375, 0.19921875, 0.256591796875, 0.035400390625, 0.161376953125, 0.2408447265625, 0.05804443359375, 0.181884765625, 0.031158447265625, 0.18505859375, 0.011322021484375, 0.039215087890625, 0.12890625, 0.0931396484375, 0.011566162109375, 0....
a165bca56220a3370ff0669d19cb053a6345a00b
subsection
74
83
Wired spanning forest oriented toward a root
Recall that X_n denotes the walker^{\prime }s location and \rho _n denotes the rotor configuration at the n-th step of the RWLM. }}}\begin{} Let (G,c) be a weighted lattice graph in \mathbb {R}^d, and suppose that the mechanism of the RWLM satisfies \ref {item: TR} and \ref {item: ELL}. Let \pi be any tail trivial stat...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 85763, 1193, 454, 19, 8, 157, 35691, 56, 114654, 31913, 497, 101, 2062, 1290, 180346, 653, 927, 29954, 78691, 37150, 10842, 724, 238, 186, 57888, 10495, 24494, 41382, 71, 139124, 191619, 40407, 3387, 29087, 217, 14107, 56051, 1434, 2499, ...
[ 0.102294921875, 0.10589599609375, 0.11279296875, 0.1656494140625, 0.0906982421875, 0.1177978515625, 0.1651611328125, 0.16748046875, 0.1165771484375, 0.1439208984375, 0.186767578125, 0.007904052734375, 0.1005859375, 0.1292724609375, 0.1519775390625, 0.0714111328125, 0.0428466796875, ...
04034fce00fa5e18bcb4abf158ec9bdbb9e00752
subsection
75
83
Wired spanning forest oriented toward a root
It follows from Definition REF and that V_i=& X_{i+1}- X_{i} = \sum _{\mathbf {y}\in {N}(\operatorname{\mathbf {0}})} \mathbb {1}\lbrace \rho _{i}(X_i)-X_i=\mathbf {y}\rbrace \, Y_{\mathbf {y},i}, where Y_{\mathbf {y},i} is a random variable on neighbors of the origin sampled from p_{\operatorname{\mathbf {0}}}(\math...
{ "cite_spans": [] }
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 1650, 28960, 7, 1295, 155455, 9069, 919, 136, 450, 310, 454, 14, 1369, 1230, 1193, 21748, 9, 2203, 41872, 11832, 150598, 53, 73, 839, 206469, 11627, 125458, 2389, 5125, 418, 99407, 42, 497, 1542, 990, 83, 96759, 77336, 98, 208244, 59665...
[ 0.1634521484375, 0.140380859375, 0.02410888671875, 0.050537109375, 0.201171875, 0.213134765625, 0.290771484375, 0.03179931640625, 0.06048583984375, 0.284912109375, 0.165283203125, 0.2376708984375, 0.1014404296875, 0.1058349609375, 0.1749267578125, 0.209228515625, 0.1358642578125, 0...
ccd85d0eda2ab18f920a3c389d997c83ffe64e65
subsection
76
83
Wired spanning forest oriented toward a root
Since \mu _{\operatorname{\mathbf {0}}} is a stationary distribution of the mechanism at \operatorname{\mathbf {0}} by , it then follows that: \lim _{n \rightarrow \infty } \frac{1}{n}\sum _{i=0}^{n-1} \mathbb {E}\mathopen {}\mathclose {\left[V_i V_i^\top \mid {F}_i\right]} =&\sum _{\mathbf {y}^{\prime } \in {N}(\oper...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1002/rsa.20747", "end": 1940, "openalex_id": "https://openalex.org/W2962799388", "raw": "Wilfried Huss, Lionel Levine, and Ecaterina Sava-Huss, Interpolating between random walk and rotor walk, Random structures & algorithms 52 (2018), n...
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 66016, 41872, 561, 206469, 11627, 150598, 2389, 8152, 83, 10, 29398, 6635, 113068, 191619, 99, 6, 24854, 125458, 47391, 390, 28960, 7, 3141, 19, 54969, 118201, 46632, 939, 51912, 132076, 418, 11832, 101, 14, 145407, 8353, 5759, 5125, 10666,...
[ 0.0029296875, 0.00921630859375, 0.2130126953125, 0.16552734375, 0.113037109375, 0.1588134765625, 0.1319580078125, 0.007720947265625, 0.06298828125, 0.012939453125, 0.1888427734375, 0.0966796875, 0.251220703125, 0.2296142578125, 0.06695556640625, 0.007720947265625, 0.008392333984375, ...
90708ffb09dfea983a895bf3aad18bf90a64375c
subsection
77
83
Unoriented wired spanning forest
We begin by defining the unoriented wired spanning forest and refer to and for a detailed discussion on this topic.Recall that G:=(V(G),E(G)) is a simple, connected, undirected graph that is locally finite. Let {F}:={F}(G) \subseteq 2^{E(G)} be the \sigma -algebra on the set of subgraphs of G generated by sets of the ...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1214/aop/1008956321", "end": 117, "openalex_id": "https://openalex.org/W1968188093", "raw": "Itai Benjamini, Russell Lyons, Yuval Peres, and Oded Schramm, Uniform spanning forests, Ann. Probab. 29 (2001), no. 1, 1–65.", "source_ref...
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 1401, 9842, 13204, 449, 70, 51, 49133, 297, 1439, 175884, 100, 525, 15005, 47, 136, 10, 185688, 35107, 98, 28451, 450, 527, 12, 1369, 856, 724, 647, 16, 83, 8781, 162711, 80581, 6, 41382, 4000, 538, 94418, 13, 10842, 919, 8152, 24854,...
[ 0.003021240234375, 0.072021484375, 0.199951171875, 0.05426025390625, 0.05560302734375, 0.1590576171875, 0.25634765625, 0.1495361328125, 0.2188720703125, 0.250244140625, 0.19287109375, 0.203125, 0.06341552734375, 0.0062255859375, 0.006072998046875, 0.00616455078125, 0.030731201171875,...
6b8cee8a01d9d5192d7b83fbe01380149b138b27
subsection
78
83
Unoriented wired spanning forest
The graph G_n is the undirected graph obtained from G by identifying all the vertices of V(G)\setminus W_n to a single vertex z_n and removing loops and extra multiple edges that are formed. The conductance c_n: E(G_n) \rightarrow \mathbb {R}_{>0} is defined byc_n\lbrace x,y\rbrace := {\mathopen {}\mathclose {\left\lbr...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1017/9781316672815", "end": 1393, "openalex_id": "https://openalex.org/W121824094", "raw": "Russell Lyons and Yuval Peres, Probability on trees and networks, Cambridge Series in Statistical and Probabilistic Mathematics, vol. 42, Cambrid...
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 581, 6, 41382, 527, 454, 19, 83, 70, 51, 80581, 297, 113054, 1295, 135812, 214, 756, 95378, 5170, 310, 724, 16, 3509, 34731, 7, 601, 47, 10, 11001, 493, 24371, 97, 49146, 6496, 40956, 136, 4173, 48716, 121303, 450, 621, 100, 4806, 5...
[ 0.06951904296875, 0.009735107421875, 0.286865234375, 0.1917724609375, 0.155029296875, 0.1859130859375, 0.110595703125, 0.035400390625, 0.0927734375, 0.21044921875, 0.120361328125, 0.130126953125, 0.05908203125, 0.1263427734375, 0.009429931640625, 0.074951171875, 0.1422119140625, 0....
1fc0e2b1e2f8e4cc859cc01b362022c6ee0bc48f
subsection
79
83
Unoriented wired spanning forest
Then, for any tail event {B}\in {F}, we have \operatorname{\mathsf {{WSF}}}[{B}] \in \lbrace 0,1\rbrace .This tail triviality will be useful for us in proving ergodicity of the scenery process for random walks with local memory later in Section .We begin by defining the unoriented wired spanning forest and refer to and...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1214/aop/1008956321", "end": 363, "openalex_id": "https://openalex.org/W1968188093", "raw": "Itai Benjamini, Russell Lyons, Yuval Peres, and Oded Schramm, Uniform spanning forests, Ann. Probab. 29 (2001), no. 1, 1–65.", "source_ref...
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 47009, 100, 2499, 46741, 19732, 571, 919, 765, 206469, 11627, 125458, 420, 52143, 73, 99407, 72966, 1927, 686, 134393, 80234, 502, 72, 112725, 60089, 28302, 9433, 96759, 35691, 678, 4000, 98323, 140978, 9842, 13204, 449, 70, 51, 49133, 297,...
[ 0.02203369140625, 0.191650390625, 0.1177978515625, 0.2091064453125, 0.19091796875, 0.125, 0.139404296875, 0.0633544921875, 0.150390625, 0.095703125, 0.05889892578125, 0.145263671875, 0.0726318359375, 0.034698486328125, 0.1185302734375, 0.1749267578125, 0.02276611328125, 0.062622070...
451cbc2324c4fd558183ad329f3a920ab7968ee1
subsection
80
83
Unoriented wired spanning forest
Let (W_n)_{n \ge 0} be a sequence of finite, connected subsets of V(G) such that\bigcup _{n \ge 0}W_n=V(G); and W_n \subseteq W_{n+1} for all n \ge 0.The wired exhaustion of G is the sequence of electrical networks (G_n, c_n)_{n \ge 0} defined as follows. The graph G_n is the undirected graph obtained from G by identi...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1017/9781316672815", "end": 1650, "openalex_id": "https://openalex.org/W121824094", "raw": "Russell Lyons and Yuval Peres, Probability on trees and networks, Cambridge Series in Statistical and Probabilistic Mathematics, vol. 42, Cambrid...
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 10842, 1456, 454, 19, 16, 6, 429, 757, 186, 944, 3956, 111, 94418, 13, 162711, 1614, 3509, 7, 310, 724, 6044, 32976, 33874, 24854, 41872, 8152, 856, 3142, 601, 22144, 59155, 864, 21748, 100, 756, 653, 3957, 1439, 297, 146353, 1830, 52...
[ 0.07354736328125, 0.109619140625, 0.06695556640625, 0.1116943359375, 0.005126953125, 0.00506591796875, 0.07781982421875, 0.1329345703125, 0.012939453125, 0.099609375, 0.005218505859375, 0.04925537109375, 0.1727294921875, 0.00750732421875, 0.175048828125, 0.1966552734375, 0.1166992187...
b4a4b011c7f637a199162094d8cd583f531b9784
subsection
81
83
Unoriented wired spanning forest
For any subset K \subseteq E(G), let {F}(K) \subseteq {F} denote the \sigma -algebra of events that depend only on K. An event {B}\in {F} is a tail event if {B}\in {F}(E \setminus K) for all finite K \subseteq E.Theorem 5.6 () Let (G,c) be an electrical network. Then, for any tail event {B}\in {F}, we have \operatorna...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1017/9781316672815", "end": 263, "openalex_id": "https://openalex.org/W121824094", "raw": "Russell Lyons and Yuval Peres, Probability on trees and networks, Cambridge Series in Statistical and Probabilistic Mathematics, vol. 42, Cambridg...
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 2499, 1614, 3509, 341, 22144, 59155, 864, 241, 724, 2633, 919, 605, 8, 48345, 20561, 192, 289, 429, 2844, 111, 47353, 56566, 4734, 98, 19732, 571, 73, 83, 10, 46741, 2174, 647, 34731, 756, 94418, 3957, 58391, 124525, 10842, 39108, 33120...
[ 0.058380126953125, 0.15234375, 0.207763671875, 0.136474609375, 0.102294921875, 0.17578125, 0.052642822265625, 0.1287841796875, 0.110107421875, 0.051025390625, 0.188720703125, 0.1221923828125, 0.0660400390625, 0.1094970703125, 0.1644287109375, 0.1771240234375, 0.002838134765625, 0.1...
368180be92ba4e66eb671230d764bb05323d3a38
subsection
82
83
Further Questions
We conclude with a few natural questions.By Theorem , if the RWLM starts at the origin and satisfies , and , then the trajectory of the walker scales to a standard Brownian motion for almost every initial rotor configuration picked from the distribution {{\operatorname{\mathsf {{WSF}}}}^+_{\operatorname{\mathbf {0}}}(G...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1002/rsa.20747", "end": 1405, "openalex_id": "https://openalex.org/W2962799388", "raw": "Wilfried Huss, Lionel Levine, and Ecaterina Sava-Huss, Interpolating between random walk and rotor walk, Random structures & algorithms 52 (2018), n...
10.1007/s10955-021-02791-5
1809.04710
Random walks with local memory
[ "Swee Hong Chan", "Lila Greco", "Lionel Levine", "Peter Li" ]
[ "math.PR" ]
2,018
en
Mathematics
[ 103876, 678, 10846, 6083, 17582, 75358, 581, 58391, 6, 2174, 78691, 37150, 4034, 99, 59665, 136, 40407, 3387, 4, 70, 182418, 30675, 111, 35691, 56, 105994, 7, 47, 5570, 40218, 3378, 78112, 39555, 11907, 61475, 2062, 1290, 180346, 171739, ...
[ 0.1138916015625, 0.012115478515625, 0.07080078125, 0.1474609375, 0.1319580078125, 0.0187835693359375, 0.03582763671875, 0.187255859375, 0.013427734375, 0.044769287109375, 0.2044677734375, 0.257568359375, 0.1361083984375, 0.105224609375, 0.2242431640625, 0.013427734375, 0.161010742187...
28b7e93b6d57653af0c143dd7d38fa1dc24ed93c
abstract
0
53
Abstract
It is commonly believed that datacenter networking software must sacrifice generality to attain high performance. The popularity of specialized distributed systems designed specifically for niche technologies such as RDMA, lossless networks, FPGAs, and programmable switches testifies to this belief. In this paper, we s...
{ "cite_spans": [] }
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 1650, 83, 39210, 538, 18822, 71, 450, 2053, 30090, 172532, 10975, 8110, 209201, 4537, 2481, 47, 99, 25500, 11192, 23718, 5, 5700, 5361, 29367, 15917, 76519, 82775, 183037, 100, 6445, 13, 118299, 6, 40756, 8218, 86669, 9393, 33120, 7, 5100...
[ 0.025177001953125, 0.044830322265625, 0.0843505859375, 0.029052734375, 0.11474609375, 0.0288238525390625, 0.0193634033203125, 0.11767578125, 0.2103271484375, 0.181884765625, 0.226318359375, 0.0692138671875, 0.13720703125, 0.179443359375, 0.1077880859375, 0.031951904296875, 0.02897644...
53aeb7cc287510ccc75dbb1708675d1196116e29
subsection
1
53
Body
0pt1.5ex plus 1ex minus .2ex1.3ex plus .2exitemsep=0pt,parsep=0pt,nosepdetect-alllanguage=C++, basicstyle=, keywordstyle=, breaklines=true, commentstyle=emph = string, u8, u64, Msg, Func, Context, Request, Response, RequestHandlerCb, Cont, EventType, ErrorType, TxDesc, sizet, pkthdrt, emphstyle =section§213 subsection§...
{ "cite_spans": [] }
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 757, 6328, 42876, 3355, 1001, 106, 26948, 6, 5, 81046, 304, 23399, 195, 70637, 145407, 4, 2500, 157, 3667, 24762, 9, 5584, 196651, 1369, 441, 37223, 62822, 40946, 166117, 36356, 59801, 9774, 13, 6868, 11727, 2203, 79315, 75, 1019, 13307, ...
[ 0.1094970703125, 0.12451171875, 0.1513671875, 0.18359375, 0.1744384765625, 0.0220489501953125, 0.1973876953125, 0.01275634765625, 0.02972412109375, 0.0787353515625, 0.031494140625, 0.155517578125, 0.07940673828125, 0.206787109375, 0.097412109375, 0.0130615234375, 0.04193115234375, ...
59cedd1df26eef93eef3237be0685c14a67083e4
subsection
2
53
Introduction
“Using performance to justify placing functions in a low-level subsystem must be done carefully. Sometimes, by examining the problem thoroughly, the same or better performance can be achieved at the high level.”— End-to-end Arguments in System DesignSqueezing the best performance out of modern, high-speed datacenter ne...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 649, "openalex_id": "https://openalex.org/W2129554014", "raw": "C. Mitchell, Y. Geng, and J. Li. Using one-sided RDMA reads to build a fast, CPU-efficient key-value store. In Proc. USENIX Annual Technical Conference, San Jose, CA,...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 1062, 6953, 23718, 47, 1660, 40383, 22466, 214, 32354, 7, 23, 10, 27226, 9, 67919, 1614, 16751, 8110, 186, 16940, 197918, 160193, 4, 390, 42276, 14653, 70, 2967, 538, 5701, 11522, 831, 69307, 71, 99, 11192, 17366, 2451, 18878, 188, 3611...
[ 0.103271484375, 0.0303955078125, 0.2626953125, 0.0404052734375, 0.10546875, 0.07049560546875, 0.1439208984375, 0.04034423828125, 0.149169921875, 0.04046630859375, 0.04022216796875, 0.040283203125, 0.1190185546875, 0.040435791015625, 0.1578369140625, 0.128662109375, 0.1552734375, 0....
194c7aa4d27e3a2c7d992706cdd01d38e96b98a5
subsection
3
53
Introduction
For example, in our two-layer testbed that resembles real deployments, each switch has 12 of dynamic buffer, while the BDP is only 19.eRPC (efficient RPC) is available at https://github.com/efficient/eRPC. Our research contributions are:We describe the design and implementation of a high-performance RPC library for dat...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 1547, "openalex_id": "", "raw": "C implementation of the Raft consensus protocol. https://github.com/willemt/raft, 2018g.", "source_ref_id": "dad7480d74f3d869d1addc948dfbf14b622be693", "start": 1404 }, { ...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 1326, 27781, 4, 23, 2446, 6626, 5259, 56, 3034, 13482, 3332, 195, 13566, 2773, 8, 13158, 53, 9035, 12638, 101089, 1556, 427, 111, 84079, 373, 18234, 12960, 154244, 4734, 6529, 13, 1052, 12233, 24500, 45964, 627, 16, 83, 19882, 99, 3975,...
[ 0.032318115234375, 0.069580078125, 0.0338134765625, 0.03363037109375, 0.012664794921875, 0.1007080078125, 0.1390380859375, 0.0199432373046875, 0.16259765625, 0.1650390625, 0.08160400390625, 0.1282958984375, 0.008331298828125, 0.125244140625, 0.0230255126953125, 0.1290283203125, 0.033...
e419470f518060c17ad6fdc3b9074f0db99504a6
subsection
4
53
Background and motivation
We first discuss aspects of modern datacenter networks relevant to eRPC. Next, we discuss limitations of existing networking software that underlie the need for eRPC.
{ "cite_spans": [] }
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 1401, 5117, 45252, 128746, 5744, 2053, 30090, 33120, 29191, 47, 28, 1052, 12233, 5, 4997, 642, 205969, 7, 144573, 172532, 10975, 1379, 6387, 3871, 100 ]
[ 0.040283203125, 0.08837890625, 0.1441650390625, 0.1279296875, 0.163818359375, 0.0933837890625, 0.22216796875, 0.1817626953125, 0.1329345703125, 0.033447265625, 0.10736083984375, 0.15478515625, 0.29736328125, 0.02587890625, 0.0692138671875, 0.0167388916015625, 0.200927734375, 0.0302...
3226c44573c1978254f78f70cc13be6157af0592
subsection
5
53
High-speed datacenter networking
Modern datacenter networks provide tens of Gbps per-port bandwidth and a few microseconds round-trip latency . They support polling-based network I/O from userspace, eliminating interrupts and system call overhead from the datapath , . eRPC uses userspace networking with polling, as in most prior high-performance netwo...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1145/2785956.2787484", "end": 110, "openalex_id": "https://openalex.org/W2036003010", "raw": "Y. Zhu, H. Eran, D. Firestone, C. Guo, M. Lipshteyn, Y. Liron, J. Padhye, S. Raindel, M. H. Yahia, and M. Zhang. Congestion control for large-s...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 18799, 2053, 30090, 33120, 7, 22691, 93396, 111, 527, 275, 6423, 117, 6982, 8753, 146984, 136, 10, 10846, 11948, 191633, 68807, 59623, 19161, 2408, 10660, 8060, 160, 30319, 77007, 87, 64, 670, 1295, 38937, 65421, 27169, 1916, 206735, 5426, ...
[ 0.193359375, 0.1356201171875, 0.2366943359375, 0.202880859375, 0.033447265625, 0.157958984375, 0.1256103515625, 0.0258026123046875, 0.003997802734375, 0.02557373046875, 0.1290283203125, 0.0958251953125, 0.1805419921875, 0.10400390625, 0.18115234375, 0.019622802734375, 0.0000915527343...
c1f97310fc582713bd1dbe0cf8362fb9a218341e
subsection
6
53
Lossless fabrics.
Lossless packet delivery is a link-level feature that prevents congestion-based packet drops. For example, PFC for Ethernet prevents a link's sender from overflowing the receiver's buffer by using pause frames. Some datacenter operators, including Microsoft, have deployed PFC at scale. This was done primarily to suppor...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1145/2785956.2787484", "end": 400, "openalex_id": "https://openalex.org/W2036003010", "raw": "Y. Zhu, H. Eran, D. Firestone, C. Guo, M. Lipshteyn, Y. Liron, J. Padhye, S. Raindel, M. H. Yahia, and M. Zhang. Congestion control for large-s...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 3731, 7, 9393, 43824, 126, 117989, 83, 3126, 67919, 60213, 56282, 158, 141933, 77007, 36069, 27781, 436, 27529, 100, 174580, 59146, 1295, 645, 118664, 53299, 373, 18234, 76968, 2674, 31384, 2053, 30090, 39933, 26719, 7244, 765, 13158, 105994,...
[ 0.1787109375, 0.1177978515625, 0.2447509765625, 0.156005859375, 0.10076904296875, 0.251708984375, 0.0811767578125, 0.1219482421875, 0.198974609375, 0.1861572265625, 0.204833984375, 0.06597900390625, 0.2059326171875, 0.1048583984375, 0.1634521484375, 0.03704833984375, 0.1185302734375,...
4d37448f76c3e42611cdac86386ef1214477e27d
subsection
7
53
Switch buffer
The increase in datacenter bandwidth has been accompanied by a corresponding decrease in round-trip time (RTT), resulting in a small BDP. Switch buffers have grown in size, to the point where “shallow-buffered” switches that use SRAM for buffering now provide tens of megabytes of shared buffer. Much of this buffer is d...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 717, "openalex_id": "", "raw": "M. Alizadeh, S. Yang, M. Sharif, S. Katti, N. McKeown, B. Prabhakar, and S. Shenker. pfabric: Minimal near-optimal datacenter transport. In Proc. ACM SIGCOMM, Hong Kong, China, Aug. 2013.", "s...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 581, 51312, 2053, 30090, 8753, 146984, 1556, 2809, 1030, 277, 46648, 297, 10, 42518, 214, 227204, 23, 68807, 59623, 1733, 20503, 618, 247, 16750, 19336, 154244, 5, 106101, 373, 18234, 7, 765, 55993, 19, 13267, 4, 47, 70, 2420, 41566, 9,...
[ 0.002349853515625, 0.171630859375, 0.1085205078125, 0.2176513671875, 0.111572265625, 0.2052001953125, 0.028656005859375, 0.013153076171875, 0.0289459228515625, 0.1744384765625, 0.08642578125, 0.0301513671875, 0.0300750732421875, 0.0306396484375, 0.0301513671875, 0.1781005859375, 0.11...
30a8675c2c7834eba5be7efff76cfccc9acd0cab
subsection
8
53
Limitations of existing options
Two reasons underlie our choice to design a new general-purpose RPC system for datacenter networks: First, existing datacenter networking software options sacrifice performance or generality, preventing unmodified applications from using the network efficiently. Second, co-designing storage software with the network is...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.5555/2616448.2616493", "end": 838, "openalex_id": "https://openalex.org/W1424350945", "raw": "E. Jeong, S. Woo, M. Jamshed, H. Jeong, S. Ihm, D. Han, and K. Park. mTCP: A highly scalable user-level TCP stack for multicore systems. In Pro...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 32964, 89397, 1379, 6387, 2446, 44126, 4331, 3525, 4537, 4717, 78381, 627, 12233, 5426, 100, 2053, 30090, 33120, 12, 23972, 4, 144573, 172532, 10975, 50717, 209201, 23718, 707, 56282, 214, 51, 13415, 47314, 86685, 1295, 70, 93766, 538, 5, ...
[ 0.1292724609375, 0.1334228515625, 0.103271484375, 0.0703125, 0.0201263427734375, 0.0966796875, 0.191162109375, 0.0689697265625, 0.185302734375, 0.1746826171875, 0.017364501953125, 0.1517333984375, 0.291259765625, 0.1959228515625, 0.044891357421875, 0.084228515625, 0.1949462890625, ...
6ba0cf5a6c7fe4f5c7c9b297ce87f001d114aa4f
subsection
9
53
Drawbacks of specialization
Co-designing distributed systems with network hardware is a well-known technique to improve performance. Co-design with RDMA is popular, with numerous examples from key-value stores , , , , , state machine replication , and transaction processing systems , , , . Programmable switches allow in-network optimizations such...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 262, "openalex_id": "https://openalex.org/W2129554014", "raw": "C. Mitchell, Y. Geng, and J. Li. Using one-sided RDMA reads to build a fast, CPU-efficient key-value store. In Proc. USENIX Annual Technical Conference, San Jose, CA,...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 1311, 9, 34198, 214, 15917, 3674, 76519, 678, 33120, 76407, 83, 5299, 69723, 19, 61353, 47, 52295, 23718, 40756, 8218, 5700, 4, 183851, 27781, 7, 22799, 27494, 13, 4343, 6, 11341, 36279, 456, 182867, 124161, 9433, 27958, 2886, 101089, 90,...
[ 0.2269287109375, 0.0579833984375, 0.326171875, 0.046722412109375, 0.24853515625, 0.0655517578125, 0.1912841796875, 0.1463623046875, 0.1605224609375, 0.2298583984375, 0.028900146484375, 0.05572509765625, 0.1199951171875, 0.02239990234375, 0.1927490234375, 0.021820068359375, 0.17749023...
28634e8da42f02877ece61bcc0dfa7d3c5620a42
subsection
10
53
eRPC overview
We provide an overview of eRPC's API and threading model below. In these aspects, eRPC is similar to existing high-performance RPC systems like Mellanox's Accelio  and FaRM. eRPC's threading model differs in how we sometimes run long-running RPC handlers in “worker” threads (§ REF ).eRPC implements RPCs on top of a tra...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 434, "openalex_id": "", "raw": "Mellanox Accelio. http://www.accelio.org, 2017b.", "source_ref_id": "d88805ab22214fbff4355535f5e583df6039e955", "start": 64 } ] }
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 22691, 645, 22751, 28, 1052, 12233, 61687, 136, 86997, 214, 3299, 35064, 128746, 83, 21373, 47, 144573, 11192, 5037, 627, 76519, 1884, 95769, 66, 17574, 29899, 54638, 3036, 25656, 7, 129927, 3642, 642, 68018, 11675, 4989, 9, 16428, 35863, ...
[ 0.0699462890625, 0.0667724609375, 0.14013671875, 0.110107421875, 0.18505859375, 0.30029296875, 0.23876953125, 0.036712646484375, 0.2037353515625, 0.12353515625, 0.22998046875, 0.123046875, 0.0699462890625, 0.06878662109375, 0.166259765625, 0.0169830322265625, 0.1431884765625, 0.064...
767c5f4026d055c9aa18b685904607a885c6b1dc
subsection
11
53
RPC API
RPCs execute at most once, and are asynchronous to avoid stalling on network round trips; intra-thread concurrency is provided using an event loop. RPC servers register request handler functions with unique request types; clients use these request types when issuing RPCs, and get continuation callbacks on RPC completio...
{ "cite_spans": [] }
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 627, 12233, 7, 71924, 13, 99, 2684, 24145, 4, 621, 10, 12654, 101966, 10821, 47, 71864, 20974, 2069, 33120, 68807, 24745, 74, 18440, 9, 927, 39116, 108636, 982, 2408, 83, 62952, 17368, 19732, 40956, 5, 10723, 68067, 50336, 35863, 32354, ...
[ 0.187255859375, 0.323974609375, 0.1075439453125, 0.211669921875, 0.047393798828125, 0.0258636474609375, 0.1470947265625, 0.2313232421875, 0.026123046875, 0.028717041015625, 0.025177001953125, 0.08935546875, 0.1544189453125, 0.02606201171875, 0.024169921875, 0.1302490234375, 0.1228027...
ea046c1d73775d622c99a3bc10100a69d00e8b4d
subsection
12
53
Client control flow:
rpc->enqueue_request() queues a request msgbuf on a session, which is transmitted when the user runs rpc's event loop. On receiving the response, the event loop copies it to the client's response msgbuf and invokes the continuation callback.
{ "cite_spans": [] }
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 1690, 57095, 103607, 33, 944, 13388, 107, 75616, 132, 16, 41, 21049, 10, 50336, 68163, 74251, 20101, 98, 56002, 71872, 3674, 3229, 70, 38937, 127877, 19732, 40956, 217063, 57553, 71200, 23282, 23, 23253, 9454, 1363, 11782, 12620, 5 ]
[ 0.0836181640625, 0.2349853515625, 0.07513427734375, 0.0950927734375, 0.16552734375, 0.28515625, 0.1317138671875, 0.2236328125, 0.012969970703125, 0.037353515625, 0.2259521484375, 0.256103515625, 0.008544921875, 0.251220703125, 0.1072998046875, 0.1114501953125, 0.2447509765625, 0.05...
edc051d74fbed5e302a520e11570c6b2ee880f51
subsection
13
53
Server control flow:
The event loop of the Rpc that owns the server session invokes (or dispatches) a request handler on receiving a request. We allow nested RPCs, i.e., the handler need not enqueue a response before returning. It may issue its own RPCs and call enqueue_response() for the first request later when all dependencies complete.
{ "cite_spans": [] }
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 581, 19732, 40956, 111, 25765, 238, 450, 10002, 7, 10723, 56002, 23, 23253, 90, 748, 58580, 257, 17007, 10, 50336, 35863, 98, 217063, 1401, 63769, 8512, 71, 627, 12233, 13, 3871, 959, 22, 944, 13388, 57553, 8108, 30646, 1543, 31089, 117...
[ 0.040496826171875, 0.23974609375, 0.285888671875, 0.0350341796875, 0.14892578125, 0.1744384765625, 0.03411865234375, 0.1976318359375, 0.08868408203125, 0.20703125, 0.2147216796875, 0.06378173828125, 0.1922607421875, 0.05908203125, 0.015777587890625, 0.06494140625, 0.0992431640625, ...
9df0ef7f30f9f92bc95b95569929c07b6572db53
subsection
14
53
Worker threads
A key design decision for an RPC system is which thread runs an RPC handler. Some RPC systems such as RAMCloud use dispatch threads for only network I/O. RAMCloud's dispatch threads communicate with worker threads that run request handlers. At datacenter network speeds, however, inter-thread communication is expensive:...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1145/2806887", "end": 382, "openalex_id": "https://openalex.org/W2074881976", "raw": "J. Ousterhout, A. Gopalan, A. Gupta, A. Kejriwal, C. Lee, B. Montazeri, D. Ongaro, S. J. Park, H. Qin, M. Rosenblum, S. Rumble, R. Stutsman, and S. Yan...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 22799, 4331, 51957, 627, 12233, 5426, 83, 3129, 86997, 127877, 35863, 5, 31384, 76519, 6044, 237, 26959, 129222, 4527, 58580, 257, 206, 7, 100, 4734, 33120, 87, 64, 670, 127219, 67, 4488, 56, 11675, 50336, 1913, 2053, 30090, 38352, 4, 4...
[ 0.1416015625, 0.2138671875, 0.1553955078125, 0.1287841796875, 0.27783203125, 0.1802978515625, 0.0189666748046875, 0.0293426513671875, 0.2437744140625, 0.125, 0.283935546875, 0.0187530517578125, 0.042388916015625, 0.190185546875, 0.038726806640625, 0.040313720703125, 0.232666015625, ...
6a9aa8e499442352175d43e7619432119f2f3afb
subsection
15
53
Evaluation clusters
Table REF shows the clusters used in this paper. They include two types of networks (lossy Ethernet, and lossless InfiniBand), and three generations of NICs released between 2011 (CX3) and 2017 (CX5). eRPC works well on all three clusters, showing that our design is robust to NIC and network technology changes. We use ...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1145/2785956.2787484", "end": 1155, "openalex_id": "https://openalex.org/W2036003010", "raw": "Y. Zhu, H. Eran, D. Firestone, C. Guo, M. Lipshteyn, Y. Liron, J. Padhye, S. Raindel, M. H. Yahia, and M. Zhang. Congestion control for large-...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 112997, 9069, 919, 45831, 234737, 7, 11814, 903, 15122, 26698, 6626, 52895, 111, 33120, 2400, 4861, 174580, 4, 86669, 9393, 360, 48310, 150501, 247, 136, 17262, 58093, 6, 67099, 121447, 17721, 1392, 15, 1542, 21320, 505, 43317, 5, 28, 105...
[ 0.170654296875, 0.09326171875, 0.1669921875, 0.06982421875, 0.25146484375, 0.053436279296875, 0.1524658203125, 0.03619384765625, 0.09698486328125, 0.0435791015625, 0.0650634765625, 0.1668701171875, 0.03448486328125, 0.17822265625, 0.109375, 0.0850830078125, 0.2064208984375, 0.03448...
40be41dcde6dff49296543c55e7a3a9ee0aa58c9
subsection
16
53
eRPC design
Achieving eRPC's performance goals requires careful design and implementation. We discuss three aspects of eRPC's design in this section: scalability of our networking primitives, the challenges involved in supporting zero-copy, and the design of sessions. The next section discusses eRPC's wire protocol and congestion ...
{ "cite_spans": [] }
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 55391, 28, 1052, 12233, 23718, 109458, 144570, 517, 7844, 4331, 208124, 45252, 17262, 128746, 903, 40059, 117906, 41159, 2446, 172532, 128489, 127125, 75412, 8060, 45234, 137366, 104227, 11737, 1439, 13, 91363, 158, 141933, 6226, 64240, 73986, ...
[ 0.10107421875, 0.09332275390625, 0.163818359375, 0.272216796875, 0.172119140625, 0.1741943359375, 0.0845947265625, 0.06964111328125, 0.00360107421875, 0.2237548828125, 0.11181640625, 0.0679931640625, 0.1278076171875, 0.143798828125, 0.003265380859375, 0.07080078125, 0.15625, 0.0594...
73d430ffeb8b1abe9ddf4d7f3ddf17ca5fe69b76
subsection
17
53
Scalability considerations
We chose plain packet I/O instead of RDMA writes , , to send messages in eRPC. This decision is based on prior insights from our design of FaSST: First, packet I/O provides completion queues that can scalably detect received packets. Second, RDMA caches connection state in NICs, which does not scale to large clusters. ...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.5555/2616448.2616486", "end": 79, "openalex_id": "https://openalex.org/W1532546444", "raw": "A. Dragojević, D. Narayanan, O. Hodson, and M. Castro. FaRM: Fast remote memory. In Proc. 11th USENIX NSDI, Seattle, WA, Apr. 2014.", "sou...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 1401, 19667, 111719, 43824, 126, 87, 64, 670, 64457, 111, 40756, 8218, 33022, 7, 6, 47, 25379, 89914, 23, 28, 1052, 12233, 51957, 35509, 41928, 149201, 4331, 3036, 294, 8545, 87344, 21721, 41, 21049, 831, 146232, 38526, 96391, 75204, 7831...
[ 0.012542724609375, 0.1275634765625, 0.1624755859375, 0.13037109375, 0.1396484375, 0.0838623046875, 0.2230224609375, 0.1947021484375, 0.1253662109375, 0.02203369140625, 0.1500244140625, 0.2401123046875, 0.1885986328125, 0.013824462890625, 0.013641357421875, 0.0243682861328125, 0.10083...
6a6e178eb08b6b136f2944f784ce02e4b84fb0df
subsection
18
53
Packet I/O scales well
RPC systems that use RDMA writes have a fundamental scalability limitation. In these systems, clients write requests directly to per-client circular buffers in the server's memory; the server must poll these buffers to detect new requests. The number of circular buffers grows with the number of clients, limiting scalab...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 1162, "openalex_id": "", "raw": "A. Kalia, M. Kaminsky, and D. G. Andersen. FaSST: Fast, scalable and simple distributed transactions with two-sided RDMA datagram RPCs. In Proc. 12th USENIX OSDI, Savannah, GA, Nov. 2016a.", ...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 627, 12233, 76519, 4527, 40756, 8218, 33022, 7, 765, 20531, 117906, 41159, 205969, 4, 19802, 50336, 105237, 47, 117, 9, 222978, 115339, 373, 18234, 23, 70, 10723, 25, 98323, 8110, 160, 1181, 96391, 14012, 111, 55993, 17475, 194583, 89160, ...
[ 0.133544921875, 0.2332763671875, 0.1513671875, 0.114013671875, 0.18505859375, 0.2353515625, 0.22314453125, 0.07135009765625, 0.023773193359375, 0.1265869140625, 0.189697265625, 0.10498046875, 0.208740234375, 0.014190673828125, 0.141357421875, 0.1806640625, 0.062042236328125, 0.0142...
9403d51f9087b64e7e6bab7944462c66562c9883
subsection
19
53
Scalability limits of RDMA
RDMA requires NIC-managed connection state. This limits scalability because NICs have limited SRAM to cache connection state. The number of in-NIC connections may be reduced by sharing them among CPU cores, but doing so reduces performance by up to 80% .Some researchers have hypothesized that improvements in NIC hardwa...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 254, "openalex_id": "", "raw": "A. Kalia, M. Kaminsky, and D. G. Andersen. FaSST: Fast, scalable and simple distributed transactions with two-sided RDMA datagram RPCs. In Proc. 12th USENIX OSDI, Savannah, GA, Nov. 2016a.", "...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 40756, 8218, 144570, 67099, 4707, 17704, 94878, 11341, 5, 3293, 17475, 7, 117906, 41159, 6637, 6, 765, 84046, 35812, 47, 82365, 581, 14012, 111, 23, 1543, 186, 34390, 71, 390, 53371, 54940, 86039, 48052, 4, 1284, 20594, 221, 23718, 1257, ...
[ 0.1708984375, 0.27392578125, 0.1563720703125, 0.205810546875, 0.1593017578125, 0.04315185546875, 0.1890869140625, 0.156982421875, 0.03564453125, 0.041168212890625, 0.1796875, 0.091064453125, 0.208740234375, 0.1595458984375, 0.043060302734375, 0.04315185546875, 0.043243408203125, 0....
115a3ad599c590c18ac472ab36155be2f6c14956
subsection
20
53
Challenges in zero-copy transmission
eRPC uses zero-copy packet I/O to provide performance comparable to low-level interfaces such as DPDK and RDMA. This section describes the challenges involved in doing so.
{ "cite_spans": [] }
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 28, 1052, 12233, 4527, 7, 45234, 9, 137366, 43824, 126, 87, 64, 670, 47, 22691, 23718, 215667, 27226, 67919, 101758, 6044, 72445, 45461, 136, 40756, 8218, 3293, 40059, 98363, 70, 127125, 75412, 20594, 221 ]
[ 0.1060791015625, 0.1671142578125, 0.30322265625, 0.1651611328125, 0.0655517578125, 0.2205810546875, 0.0255279541015625, 0.2242431640625, 0.141357421875, 0.0797119140625, 0.034332275390625, 0.1878662109375, 0.1700439453125, 0.056915283203125, 0.1236572265625, 0.20068359375, 0.16772460...
912b4fc5e4daa48f74e107671cfa112e882a5bcb
subsection
21
53
Message buffer layout
eRPC provides DMA-capable message buffers to applications for zero-copy transfers. A msgbuf holds one, possibly multi-packet message. It consists of per-packet headers and data, arranged in a fashion optimized for small single-packet messages (Figure REF ). Each eRPC packet has a header that contains the transport head...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 789, "openalex_id": "", "raw": "A. Kalia, M. Kaminsky, and D. G. Andersen. Design guidelines for high-performance RDMA systems. In Proc. USENIX Annual Technical Conference, Denver, CO, June 2016b.", "source_ref_id": "5e2cc3d...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 28, 1052, 12233, 87344, 391, 8218, 15644, 2886, 26008, 373, 18234, 7, 47, 86685, 100, 45234, 137366, 12302, 62, 68163, 74251, 20101, 16401, 1632, 144681, 6024, 29102, 126, 58055, 117, 10336, 1314, 136, 2053, 137356, 54543, 15572, 19336, 110...
[ 0.107666015625, 0.18115234375, 0.30908203125, 0.1151123046875, 0.04595947265625, 0.19140625, 0.1317138671875, 0.0599365234375, 0.143798828125, 0.1270751953125, 0.228759765625, 0.028167724609375, 0.050079345703125, 0.1729736328125, 0.047454833984375, 0.1719970703125, 0.185791015625, ...
de8655e040b8b4fe79c133627cf0eaab24fa8f8e
subsection
22
53
Message buffer ownership
Since eRPC transfers packets directly from application-owned msgbufs, msgbuf references must never be used by eRPC after msgbuf ownership is returned to the application. In this paper, we discuss msgbuf ownership issues for only clients; the process is similar but simpler for the server, since eRPC's servers are passiv...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1145/2619239.2626299", "end": 1724, "openalex_id": "https://openalex.org/W2140101134", "raw": "A. Kalia, M. Kaminsky, and D. G. Andersen. Using RDMA efficiently for key-value services. In Proc. ACM SIGCOMM, Chicago, IL, Aug. 2014.", ...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 28, 1052, 12233, 12302, 43824, 7831, 105237, 38415, 9, 157272, 68163, 177, 978, 17447, 4, 74251, 20101, 91067, 8306, 186, 11814, 101785, 16070, 83, 176377, 47, 70, 360, 15122, 37348, 100, 4734, 19802, 9433, 21373, 1284, 8781, 42, 10723, 7...
[ 0.074951171875, 0.150634765625, 0.2890625, 0.13427734375, 0.110107421875, 0.014923095703125, 0.0699462890625, 0.140380859375, 0.008270263671875, 0.1617431640625, 0.1187744140625, 0.04827880859375, 0.1197509765625, 0.15185546875, 0.0084228515625, 0.1300048828125, 0.252685546875, 0.2...
71f0ca270ada6773c9e64aafa19d91ea2d70fc60
subsection
23
53
Zero-copy request processing
Zero-copy reception is harder than transmission: To provide a contiguous request msgbuf to the request handler at the server, we must strip headers from received packets, and copy only application data to the target msgbuf. However, we were able to provide zero-copy reception for our common-case workload consisting of ...
{ "cite_spans": [] }
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 83947, 9, 137366, 116816, 83, 182, 820, 3501, 179965, 717, 22691, 22832, 12282, 50336, 68163, 74251, 20101, 47, 35863, 10723, 8110, 43613, 10336, 1314, 75204, 43824, 43658, 4734, 38415, 2053, 30388, 33306, 19048, 45234, 100, 39210, 58437, 448...
[ 0.2164306640625, 0.07489013671875, 0.266845703125, 0.24365234375, 0.043701171875, 0.145751953125, 0.1513671875, 0.04302978515625, 0.1895751953125, 0.020751953125, 0.134765625, 0.0733642578125, 0.161376953125, 0.2135009765625, 0.050445556640625, 0.07769775390625, 0.2149658203125, 0....
150a008138067dff5c440b96cef9b9b7d67ee26c
subsection
24
53
Sessions
Each session maintains multiple outstanding requests to keep the network pipe full. Concurrently requests on a session can complete out-of-order with respect to each other. This avoids blocking dispatch-mode RPCs behind a long-running worker-mode RPC. We support a constant number of concurrent requests (default = 8) pe...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 465, "openalex_id": "", "raw": "RDMAmojo - blog on RDMA technology and programming by Dotan Barak. http://www.rdmamojo.com/2013/01/12/ibv_modify_qp/, 2018c.", "source_ref_id": "c5b8a6780a44391b0510f45972544e624877231d", ...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 98423, 56002, 76104, 7, 48716, 216348, 50336, 47, 13695, 70, 33120, 137158, 4393, 5, 1657, 163812, 538, 98, 10, 831, 28484, 1810, 9, 4390, 80596, 678, 15072, 12638, 3789, 3293, 71864, 46389, 58580, 257, 206, 61170, 627, 12233, 50155, 4989...
[ 0.1412353515625, 0.2308349609375, 0.1610107421875, 0.065673828125, 0.1510009765625, 0.207763671875, 0.23974609375, 0.027618408203125, 0.1353759765625, 0.031890869140625, 0.132080078125, 0.1558837890625, 0.09619140625, 0.0294952392578125, 0.048797607421875, 0.1912841796875, 0.01428222...
ceb9a1df18122e7ab8bde0abfa34b7b797920d9f
subsection
25
53
Session credits
eRPC limits the number of unacknowledged packets on a session for two reasons. First, to avoid dropping packets due to an empty RQ with no descriptors, the number of packets that may be sent to an Rpc must not exceed the size of its RQ (|RQ|). Because each session sends packets independently of others, we first limit t...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1145/3230543.3230557", "end": 883, "openalex_id": "https://openalex.org/W2809353470", "raw": "R. Mittal, A. Shpiner, A. Panda, E. Zahavi, A. Krishnamurthy, S. Ratnasamy, and S. Shenker. Revisiting network support for RDMA. In Proc. ACM S...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 28, 1052, 12233, 17475, 7, 14012, 220, 238, 69723, 181606, 43824, 7831, 98, 10, 56002, 100, 6626, 89397, 71864, 36069, 4743, 201505, 627, 2737, 110, 104, 121707, 1543, 9325, 47, 142, 25765, 8110, 204839, 6863, 58745, 88949, 12638, 25379, ...
[ 0.0960693359375, 0.1593017578125, 0.282470703125, 0.230712890625, 0.054931640625, 0.101318359375, 0.088623046875, 0.1854248046875, 0.155029296875, 0.1500244140625, 0.1583251953125, 0.1258544921875, 0.068359375, 0.03900146484375, 0.1962890625, 0.018524169921875, 0.115966796875, 0.11...
cf070b86af2a839a1dd1565f5c8cf2a33ab28b9f
subsection
26
53
Session scalability
eRPC's scalability depends on the user's desired value of C, and the number and size of RQs that the NIC and host can effectively support. Lowering C increases scalability, but reduces session throughput by restricting the session's packet window. Small values of C (e.g., C = 1) should be used in applications that (a) ...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 1608, "openalex_id": "", "raw": "A. Dragojević, D. Narayanan, E. B. Nightingale, M. Renzelmann, A. Shamis, A. Badam, and M. Castro. No compromises: Distributed transactions with consistency, availability, and performance. In Proc....
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 28, 1052, 12233, 117906, 41159, 56566, 98, 70, 38937, 25, 7, 104851, 71, 34292, 111, 313, 4, 136, 13267, 627, 2737, 450, 67099, 27980, 831, 191984, 8060, 61187, 4209, 51312, 1284, 34390, 56002, 8305, 7077, 390, 173072, 214, 43824, 126, ...
[ 0.10302734375, 0.172607421875, 0.294677734375, 0.2271728515625, 0.1839599609375, 0.143310546875, 0.047576904296875, 0.047576904296875, 0.103271484375, 0.047515869140625, 0.047637939453125, 0.144287109375, 0.047393798828125, 0.11572265625, 0.047607421875, 0.1414794921875, 0.0476684570...
49bb2bd57ca484c4fa12464253615f183332f3b7
subsection
27
53
Wire protocol
We designed a wire protocol for eRPC that is optimized for small RPCs and accounts for per-session credit limits. For simplicity, we chose a simple client-driven protocol, meaning that each packet sent by the server is in response to a client packet. A client-driven protocol has fewer “moving parts” than a protocol in ...
{ "cite_spans": [] }
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 1401, 82775, 10, 1439, 13, 91363, 28, 1052, 12233, 15572, 29367, 100, 19336, 627, 15426, 117, 169234, 22299, 17475, 134381, 642, 19667, 8781, 23282, 9, 33926, 33, 91084, 12638, 43824, 126, 9325, 10723, 57553, 62, 1556, 10846, 56, 6496, 63...
[ 0.0328369140625, 0.17041015625, 0.000396728515625, 0.1861572265625, 0.1119384765625, 0.265380859375, 0.06842041015625, 0.11083984375, 0.2384033203125, 0.1580810546875, 0.08447265625, 0.025238037109375, 0.127685546875, 0.045928955078125, 0.09228515625, 0.03497314453125, 0.152099609375...
44675c5d353c82a146b18fba240a7387c1d8af1f
subsection
28
53
Protocol messages
Figure REF shows the packets sent with C = 2 for a small single-packet RPC, and for an RPC whose request and response require three packets each. Single-packet RPCs use the fewest packets possible. The client begins by sending a window of up to C request data packets. For each request packet except the last, the server...
{ "cite_spans": [] }
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 55412, 9069, 919, 45831, 43824, 7831, 9325, 678, 313, 2203, 116, 19336, 11001, 29102, 126, 627, 12233, 50336, 136, 57553, 64209, 17262, 12638, 43609, 7, 4527, 10846, 525, 7722, 23282, 9842, 135834, 76896, 2053, 40494, 4568, 10723, 25379, 44...
[ 0.136962890625, 0.09539794921875, 0.1571044921875, 0.005401611328125, 0.16796875, 0.13623046875, 0.068359375, 0.0423583984375, 0.110107421875, 0.0657958984375, 0.10491943359375, 0.052215576171875, 0.0877685546875, 0.196533203125, 0.083251953125, 0.1190185546875, 0.248779296875, 0.1...
2251118ac576261fc969e29b300bf9d1f4a19044
subsection
29
53
Congestion control
Congestion control for datacenter networks aims to reduce switch queueing, thereby preventing packet drops and reducing RTT. Prior high-performance RPC implementations such as FaSST do not implement congestion control, and some researchers have hypothesized that doing so will substantially reduce performance . Can effe...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 311, "openalex_id": "https://openalex.org/W2620820275", "raw": "A. Dragojevic, D. Narayanan, and M. Castro. RDMA reads: To use or not to use? IEEE Data Eng. Bull., 2017.", "source_ref_id": "9fc89b4bfc9db8789f6973ec4747072b49...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 1657, 141933, 6226, 100, 2053, 30090, 33120, 131264, 47, 34390, 101089, 41, 13388, 56282, 43824, 36069, 241866, 627, 13739, 135967, 11192, 5037, 12233, 208124, 3036, 294, 8545, 959, 29479, 158, 25188, 170933, 193984, 23718, 4171, 60266, 297, ...
[ 0.142333984375, 0.271484375, 0.240234375, 0.054656982421875, 0.0860595703125, 0.18359375, 0.138427734375, 0.143798828125, 0.0284576416015625, 0.146484375, 0.17138671875, 0.0419921875, 0.07525634765625, 0.118408203125, 0.06036376953125, 0.1097412109375, 0.1356201171875, 0.0565185546...
200ad4d14d7ffd4a2874f219bba03bca1cd46419
subsection
30
53
Available options
Congestion control for high-speed datacenter networks is an evolving area of research, with two major approaches for commodity hardware: RTT-based approaches such as Timely , and ECN-based approaches such as DCQCN . Timely and DCQCN have been deployed at Google and Microsoft, respectively. We wish to use these protocol...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 215, "openalex_id": "", "raw": "R. Mittal, T. Lam, N. Dukkipati, E. Blem, H. Wassel, M. Ghobadi, A. Vahdat, Y. Wang, D. Wetherall, and D. Zats. TIMELY: RTT-based congestion control for the datacenter. In Proc. ACM SIGCOMM, London,...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 1657, 141933, 6226, 100, 11192, 175870, 2053, 30090, 33120, 83, 3784, 6496, 16128, 111, 25188, 4, 678, 6626, 13036, 51515, 90, 123633, 939, 76407, 627, 13739, 77007, 19055, 538, 6, 136, 241, 36115, 9, 237, 31455, 2737, 765, 13158, 53, 2...
[ 0.1573486328125, 0.304931640625, 0.255615234375, 0.02581787109375, 0.1021728515625, 0.1787109375, 0.1075439453125, 0.1983642578125, 0.1307373046875, 0.03497314453125, 0.08001708984375, 0.011260986328125, 0.077880859375, 0.011749267578125, 0.1376953125, 0.011505126953125, 0.0061035156...
0cbfa255ee57638d658bae07674762ab8c6c188b
subsection
31
53
Common-case optimizations
We use three optimizations for our common-case workloads. Our evaluation shows that these optimizations reduce the overhead of congestion control from 20% to 9%, and that they do not reduce the effectiveness of congestion control. The first two are based on the observation that datacenter networks are typically unconge...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 637, "openalex_id": "", "raw": "A. Roy, H. Zeng, J. Bagga, G. Porter, and A. C. Snoeren. Inside the social network's (datacenter) network. In Proc. ACM SIGCOMM, London, UK, Aug. 2015.", "source_ref_id": "b8fcdf6f79ca90f022c3...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 1401, 4527, 17262, 87235, 5256, 100, 2446, 39210, 58437, 4488, 63033, 219836, 45831, 34390, 645, 31251, 158, 141933, 6226, 1295, 12719, 47, 6, 12678, 4, 959, 70, 60266, 6626, 621, 35509, 98, 150556, 2053, 30090, 33120, 7, 205794, 51, 2271...
[ 0.03765869140625, 0.1119384765625, 0.184326171875, 0.2265625, 0.1260986328125, 0.011688232421875, 0.0257720947265625, 0.154052734375, 0.1612548828125, 0.08343505859375, 0.1849365234375, 0.1148681640625, 0.01666259765625, 0.19775390625, 0.1119384765625, 0.198486328125, 0.0908203125, ...
caa3f01727cdb0d026140d071a32ea25dbf25a85
subsection
32
53
Comparison with IRN
IRN  is a new RDMA NIC architecture designed for lossy networks, with two key improvements. First, it uses BDP flow control to limit the outstanding data per RDMA connection to one BDP. Second, it uses efficient selective acks instead of simple go-back-N for packet loss recovery.IRN was evaluated with simulated switche...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1145/3230543.3230557", "end": 91, "openalex_id": "https://openalex.org/W2809353470", "raw": "R. Mittal, A. Shpiner, A. Panda, E. Zahavi, A. Krishnamurthy, S. Ratnasamy, and S. Shenker. Revisiting network support for RDMA. In Proc. ACM SI...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 87, 50456, 83, 10, 3525, 40756, 8218, 67099, 159958, 82775, 100, 388, 4861, 33120, 7, 4, 678, 6626, 22799, 136912, 23972, 442, 4527, 154244, 86608, 6226, 47, 17475, 70, 216348, 2053, 94878, 1632, 77648, 93766, 36849, 5844, 22824, 64457, 1...
[ 0.2626953125, 0.36865234375, 0.160400390625, 0.047271728515625, 0.12158203125, 0.1611328125, 0.2252197265625, 0.20703125, 0.2352294921875, 0.1944580078125, 0.110107421875, 0.1932373046875, 0.2030029296875, 0.13134765625, 0.030731201171875, 0.0306396484375, 0.028564453125, 0.1378173...
a896b483d882a4b1c7a8981f72e94ba6058ebe29
subsection
33
53
Handling packet loss
For simplicity, eRPC treats reordered packets as losses by dropping them. This is not a major deficiency because datacenter networks typically use ECMP for load balancing, which preserves intra-flow ordering , , except during rare route churn events. Note that current RDMA NICs also drop reordered packets .On suspectin...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 251, "openalex_id": "", "raw": "J. Zhou, M. Tewari, M. Zhu, A. Kabbani, L. Poutievski, A. Singh, and A. Vahdat. WCMP: Weighted cost multipathing for improved fairness in data centers. In Proc. 9th ACM European Conference on Comput...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 134381, 939, 28, 1052, 12233, 85689, 7, 456, 80596, 297, 43824, 7831, 237, 388, 5908, 390, 36069, 10366, 2856, 959, 13036, 8, 11044, 2053, 30090, 33120, 205794, 4527, 59947, 9088, 100, 72367, 25168, 60636, 479, 86687, 18440, 118664, 12989, ...
[ 0.115234375, 0.0413818359375, 0.09442138671875, 0.1494140625, 0.286376953125, 0.133544921875, 0.003814697265625, 0.120849609375, 0.25244140625, 0.06988525390625, 0.169677734375, 0.138916015625, 0.06988525390625, 0.1246337890625, 0.0999755859375, 0.0281524658203125, 0.187744140625, ...
83d0b39cfd238f3896d60c04e6c648b7bc853216
subsection
34
53
Microbenchmarks
eRPC is implemented in 6200 SLOC of C++, excluding tests and benchmarks. We use static polymorphism to create an Rpc class that works with multiple transport types without the overhead of virtual function calls. In this section, we evaluate eRPC's latency, message rate, scalability, and bandwidth using microbenchmarks....
{ "cite_spans": [] }
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 28, 1052, 12233, 83, 29479, 297, 23, 305, 5955, 159, 124448, 313, 37223, 39041, 6238, 109921, 136, 240057, 1401, 4527, 201939, 35874, 178851, 8780, 28282, 25765, 238, 18507, 43240, 678, 48716, 6181, 52895, 15490, 31251, 111, 20513, 32354, 1...
[ 0.1199951171875, 0.1871337890625, 0.315185546875, 0.0273590087890625, 0.22021484375, 0.03680419921875, 0.0679931640625, 0.1219482421875, 0.267578125, 0.0750732421875, 0.2376708984375, 0.058349609375, 0.24658203125, 0.08099365234375, 0.015625, 0.1561279296875, 0.008575439453125, 0.1...
0aaf0b5fea74051cadd43d3c94f56aa07bb998d9
subsection
35
53
Small RPC latency
How much latency does eRPC add? Table REF compares the median latency of 32 RPCs and RDMA reads between two nodes connected to the same ToR switch. Across all clusters, eRPC is at most 800 slower than RDMA reads.eRPC's median latency on CX5 is only 2.3, showing that latency with commodity Ethernet NICs and software net...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 387, "openalex_id": "https://openalex.org/W2798641544", "raw": "X. Jin, X. Li, H. Zhang, N. Foster, J. Lee, R. Soulé, C. Kim, and I. Stoica. NetChain: Scale-free sub-RTT coordination. In Proc. 15th USENIX NSDI, Renton, WA, Apr. 20...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 11249, 5045, 19161, 2408, 28, 1052, 12233, 15190, 112997, 9069, 919, 69101, 2450, 19, 2789, 627, 136, 40756, 8218, 12301, 17721, 6626, 110, 162711, 5701, 717, 101089, 95665, 756, 234737, 83, 2684, 7607, 72803, 56, 3501, 13, 313, 1542, 758...
[ 0.00299072265625, 0.0946044921875, 0.2403564453125, 0.139404296875, 0.0843505859375, 0.1788330078125, 0.278076171875, 0.172607421875, 0.1146240234375, 0.049957275390625, 0.1121826171875, 0.126220703125, 0.1513671875, 0.08551025390625, 0.1917724609375, 0.08673095703125, 0.028442382812...
f8f3daf63fcbbc4257235b6fb0f5fc3618388b48
subsection
36
53
Small RPC rate
What is the CPU cost of providing generality in an RPC system? We compare eRPC's small message performance against FaSST RPCs, which outperform other RPC systems such as FaRM . FaSST RPCs are specialized for single-packet RPCs in a lossless network, and they do not handle congestion.We mimic FaSST's experiment setting:...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 176, "openalex_id": "", "raw": "A. Kalia, M. Kaminsky, and D. G. Andersen. FaSST: Fast, scalable and simple distributed transactions with two-sided RDMA datagram RPCs. In Proc. 12th USENIX OSDI, Savannah, GA, Nov. 2016a.", "...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 4865, 70, 86039, 11034, 101904, 4537, 2481, 627, 12233, 5426, 69101, 28, 1052, 19336, 26008, 23718, 26548, 3036, 294, 8545, 7, 4, 1810, 1264, 5037, 3789, 76519, 25656, 621, 5361, 29367, 100, 11001, 9, 29102, 126, 23, 10, 86669, 9393, 33...
[ 0.008331298828125, 0.0089111328125, 0.23876953125, 0.1590576171875, 0.125, 0.1724853515625, 0.10888671875, 0.1202392578125, 0.27685546875, 0.1175537109375, 0.1368408203125, 0.07769775390625, 0.1549072265625, 0.14990234375, 0.135498046875, 0.16552734375, 0.07305908203125, 0.15185546...
2492d212f354db61460f0d07392255232856ce08
subsection
37
53
Factor analysis.
How important are eRPC's common-case optimizations? Table REF shows the performance impact of disabling some of eRPC's common-case optimizations on CX4; other optimizations such as our single-DMA msgbuf format and unsignaled transmissions cannot be disabled easily. For our baseline, we use B = 3 and enable congestion c...
{ "cite_spans": [] }
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 5526, 28, 1052, 12233, 39210, 58437, 87235, 5256, 112997, 9069, 919, 45831, 23718, 24725, 6392, 79298, 3060, 98, 313, 1542, 617, 237, 2446, 11001, 8218, 68163, 74251, 20101, 9384, 51, 165992, 179965, 53418, 238335, 72546, 3647, 2256, 4527, ...
[ 0.1375732421875, 0.0462646484375, 0.128662109375, 0.24462890625, 0.1673583984375, 0.200927734375, 0.21630859375, 0.0745849609375, 0.1456298828125, 0.1009521484375, 0.137939453125, 0.0261688232421875, 0.175537109375, 0.1734619140625, 0.1195068359375, 0.1483154296875, 0.018783569335937...
322726f7824ef5b7390aead2c257841acb37446a
subsection
38
53
Large RPC bandwidth
We evaluate eRPC's bandwidth using a client thread that sends large messages to a remote server thread. The client sends R-byte requests and keeps one request outstanding; the server replies with a small 32 response. We use up to 8 requests, which is the largest message size supported by eRPC. We use 32 credits per ses...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 1088, "openalex_id": "", "raw": "Fast memcpy with SPDK and Intel I/OAT DMA Engine. https://software.intel.com/en-us/articles/fast-memcpy-using-spdk-and-ioat-dma-engine.", "source_ref_id": "ec25de1fb833a1430c040b47b4939807ec4...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 151575, 28, 1052, 12233, 8753, 146984, 17368, 23282, 86997, 450, 25379, 7, 21334, 89914, 47, 148814, 10723, 627, 61028, 50336, 136, 13695, 1632, 216348, 70, 156405, 90, 678, 10, 19336, 2789, 57553, 4527, 1257, 382, 4, 3129, 83, 142105, 26...
[ 0.1170654296875, 0.114501953125, 0.1884765625, 0.30908203125, 0.1593017578125, 0.2247314453125, 0.003509521484375, 0.17529296875, 0.16015625, 0.0292205810546875, 0.035430908203125, 0.0294189453125, 0.058837890625, 0.10693359375, 0.0294189453125, 0.1839599609375, 0.148681640625, 0.1...
7aba26d44f137c47519cb3f657d5030495c43f1c
subsection
39
53
Effectiveness of congestion control
We evaluate if our congestion control is successful at reducing switch queueing. We create an incast traffic pattern by increasing the number of client nodes in the previous setup (R = 8). The one server node acts as the incast victim. During an incast, queuing primarily happens at the victim's ToR switch. We use per-p...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 391, "openalex_id": "", "raw": "R. Mittal, T. Lam, N. Dukkipati, E. Blem, H. Wassel, M. Ghobadi, A. Vahdat, Y. Wang, D. Wetherall, and D. Zats. TIMELY: RTT-based congestion control for the datacenter. In Proc. ACM SIGCOMM, London,...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 151575, 2174, 158, 141933, 6226, 65771, 241866, 101089, 41, 13388, 214, 28282, 23, 55741, 83629, 103510, 118055, 23282, 110, 988, 96362, 169581, 1052, 382, 1632, 10723, 112, 27992, 237, 91519, 133698, 34, 102917, 96276, 717, 4527, 117, 29102,...
[ 0.0816650390625, 0.01263427734375, 0.09539794921875, 0.2213134765625, 0.2119140625, 0.150390625, 0.1676025390625, 0.21044921875, 0.1434326171875, 0.12841796875, 0.050628662109375, 0.07342529296875, 0.1282958984375, 0.21826171875, 0.173583984375, 0.173095703125, 0.11376953125, 0.140...
64e21e1617edbb1b31be3672790cb23e5a103d9d
subsection
40
53
Incast with background traffic.
Next, we augment the setup above to mimic an experiment from Timely : we create one additional thread at each node that is not the incast victim. These threads exchange latency-sensitive RPCs (64 request and response), keeping one RPC outstanding. During a 100-way incast, the 99th percentile latency of these RPCs is 27...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 145, "openalex_id": "", "raw": "R. Mittal, T. Lam, N. Dukkipati, E. Blem, H. Wassel, M. Ghobadi, A. Vahdat, Y. Wang, D. Wetherall, and D. Zats. TIMELY: RTT-based congestion control for the datacenter. In Proc. ACM SIGCOMM, London,...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 4997, 642, 9620, 674, 169581, 36917, 324, 21068, 28007, 1295, 19055, 538, 28282, 1632, 78301, 86997, 12638, 110, 112, 959, 23, 55741, 91519, 121122, 19161, 2408, 176302, 13, 627, 12233, 7, 13307, 50336, 136, 57553, 120260, 216348, 133698, 3...
[ 0.06768798828125, 0.019378662109375, 0.140380859375, 0.1273193359375, 0.1854248046875, 0.0653076171875, 0.0831298828125, 0.10638427734375, 0.2105712890625, 0.0160369873046875, 0.2115478515625, 0.198974609375, 0.08709716796875, 0.037445068359375, 0.1036376953125, 0.2060546875, 0.04953...
5661e5cb6bda592a964b3702c396622c669709de
subsection
41
53
Full-system benchmarks
In this section, we evaluate whether eRPC can be used in real applications with unmodified existing storage software: We build a state machine replication system using an open-source implementation of Raft , and a networked ordered key-value store using Masstree .
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 264, "openalex_id": "https://openalex.org/W2156580773", "raw": "D. Ongaro and J. Ousterhout. In search of an understandable consensus algorithm. In Proc. USENIX Annual Technical Conference, Philadelphia, PA, June 2014.", "so...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 903, 40059, 151575, 36766, 28, 1052, 12233, 831, 186, 11814, 23, 2773, 86685, 678, 51, 13415, 47314, 144573, 132096, 10975, 1401, 45367, 11341, 36279, 456, 182867, 5426, 17368, 9803, 60427, 208124, 99083, 18, 33120, 297, 12989, 22799, 27494, ...
[ 0.024444580078125, 0.08447265625, 0.122314453125, 0.051727294921875, 0.10125732421875, 0.1546630859375, 0.28955078125, 0.10662841796875, 0.032745361328125, 0.1649169921875, 0.04669189453125, 0.18408203125, 0.1851806640625, 0.0811767578125, 0.047760009765625, 0.08251953125, 0.06481933...
c8df634a3dd0760c8577c54bc46af97757a403b3
subsection
42
53
Raft over eRPC
State machine replication (SMR) is used to build fault-tolerant services. An SMR service consists of a group of server nodes that receive commands from clients. SMR protocols ensure that each server executes the same sequence of commands, and that the service remains available if servers fail. Raft  is such a protocol ...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 508, "openalex_id": "https://openalex.org/W2156580773", "raw": "D. Ongaro and J. Ousterhout. In search of an understandable consensus algorithm. In Proc. USENIX Annual Technical Conference, Philadelphia, PA, June 2014.", "so...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 22836, 36279, 456, 182867, 294, 52005, 16, 83, 11814, 47, 45367, 193857, 131145, 660, 11374, 5, 893, 159, 4516, 58055, 7, 111, 10, 21115, 10723, 110, 988, 450, 53299, 75101, 1295, 19802, 91363, 63284, 12638, 71924, 90, 70, 5701, 944, 39...
[ 0.18017578125, 0.1727294921875, 0.1922607421875, 0.275634765625, 0.117431640625, 0.31640625, 0.009307861328125, 0.054473876953125, 0.1365966796875, 0.03045654296875, 0.10888671875, 0.1749267578125, 0.1607666015625, 0.0262298583984375, 0.16015625, 0.0264739990234375, 0.0318603515625, ...
1b752c87a87c1fc03378e94189b3972cca62e301
subsection
43
53
Workloads.
We mimic NetChain and ZabFPGA's experiment setups for latency measurement: we implement a 3-way replicated in-memory key-value store, and use one client to issue PUT requests. The replicas' command logs and key-value store are stored in DRAM. NetChain and ZabFPGA use 16 keys, and 16–64 values; we use 16 keys and 64 val...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 509, "openalex_id": "", "raw": "S. Li, H. Lim, V. W. Lee, J. H. Ahn, A. Kalia, M. Kaminsky, D. G. Andersen, O. Seongil, S. Lee, and P. Dubey. Architecting to achieve a billion requests per second throughput on a single key-value s...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 1401, 324, 21068, 10086, 62467, 73, 136, 825, 275, 51006, 14849, 28007, 169581, 19161, 2408, 72350, 674, 12, 642, 29479, 5691, 7514, 143126, 3674, 23, 109005, 22799, 27494, 13, 4343, 4527, 1632, 23282, 31089, 124599, 50336, 75101, 12684, 35...
[ 0.052490234375, 0.127685546875, 0.1607666015625, 0.1334228515625, 0.221923828125, 0.1318359375, 0.087158203125, 0.14013671875, 0.1414794921875, 0.1422119140625, 0.1590576171875, 0.2301025390625, 0.1668701171875, 0.271484375, 0.1314697265625, 0.201904296875, 0.047119140625, 0.002655...
53e77a9caf88264134853574e57b6ec79d496604
subsection
44
53
Comparison with NetChain
NetChain's key assumption is that software networking adds 1–2 orders of magnitude more latency than switches . However, we have shown that eRPC adds 850, which is only around 2x higher than latency added by current programmable switches (400 ).Raft's latency over eRPC is 5.5, which is substantially lower than NetChain...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 111, "openalex_id": "https://openalex.org/W2798641544", "raw": "X. Jin, X. Li, H. Zhang, N. Foster, J. Lee, R. Soulé, C. Kim, and I. Stoica. NetChain: Scale-free sub-RTT coordination. In Proc. 15th USENIX NSDI, Renton, WA, Apr. 20...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 10086, 62467, 73, 7, 22799, 237259, 83, 10975, 172532, 15190, 304, 12989, 101668, 1286, 19161, 2408, 3501, 101089, 90, 33306, 127887, 28, 1052, 12233, 71242, 4734, 10932, 116, 425, 77546, 49814, 43581, 11720, 2661, 10700, 12248, 2480, 645, ...
[ 0.18505859375, 0.28076171875, 0.237060546875, 0.01275634765625, 0.1883544921875, 0.243896484375, 0.015625, 0.2384033203125, 0.187255859375, 0.16650390625, 0.04583740234375, 0.086669921875, 0.05218505859375, 0.1136474609375, 0.23779296875, 0.1348876953125, 0.0161285400390625, 0.2091...
2ff26922b92ac222e0e2e3b63f7f00894d1b663b
subsection
45
53
Comparison with ZabFPGA
Although ZabFPGA's SMR servers are FPGAs, the clients are commodity workstations that communicate with the FPGAs over slow kernel-based TCP. For a challenging comparison, we compare against ZabFPGA's commit latency measured at the leader, which involves only FPGAs. In addition, we consider its “direct connect” mode, wh...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 802, "openalex_id": "", "raw": "Z. István, D. Sidler, G. Alonso, and M. Vukolic. Consensus in a box: Inexpensive coordination in hardware. In Proc. 13th USENIX NSDI, Santa Clara, CA, May 2016.", "source_ref_id": "deeec81a890...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 106073, 825, 275, 51006, 14849, 159, 52005, 10723, 7, 621, 19802, 123633, 939, 4488, 41311, 127219, 678, 72803, 77924, 583, 77007, 17854, 223920, 225490, 69101, 26548, 375, 4007, 19161, 2408, 72350, 99, 57724, 83687, 4734, 80581, 37067, 13736...
[ 0.043609619140625, 0.1656494140625, 0.172119140625, 0.1561279296875, 0.1832275390625, 0.0667724609375, 0.2318115234375, 0.175048828125, 0.0264434814453125, 0.09271240234375, 0.20166015625, 0.178955078125, 0.0958251953125, 0.06298828125, 0.1358642578125, 0.1405029296875, 0.07635498046...
b6516ef0342097120279ec4c46567a698ca2358b
subsection
46
53
Masstree over eRPC
Masstree  is an ordered in-memory key-value store. We use it to implement a single-node database index that supports low-latency point queries in the presence of less performance-critical longer-running scans. This requires running scans in worker threads. We use CX3 for this experiment to show that eRPC works well on ...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1145/2168836.2168855", "end": 50, "openalex_id": "https://openalex.org/W2165663045", "raw": "Y. Mao, E. Kohler, and R. T. Morris. Cache craftiness for fast multicore key-value storage. In Proc. 7th ACM European Conference on Computer Sys...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 74227, 62600, 83, 142, 12989, 297, 23, 109005, 53, 22799, 27494, 13, 4343, 4527, 29479, 10, 11001, 157, 112, 63399, 63262, 8060, 7, 27226, 9, 19309, 27771, 6275, 41, 70, 169424, 111, 40715, 23718, 59869, 21533, 51713, 16428, 592, 44954, ...
[ 0.18896484375, 0.35791015625, 0.1033935546875, 0.0382080078125, 0.1722412109375, 0.01495361328125, 0.0494384765625, 0.10205078125, 0.0261077880859375, 0.1597900390625, 0.139404296875, 0.019989013671875, 0.194091796875, 0.090087890625, 0.108642578125, 0.0143280029296875, 0.12341308593...
993f01541a867239a79fbca83b68a833d1b8b601
subsection
47
53
RPCs.
There is a vast amount of literature on RPCs. The practice of optimizing an RPC wire protocol for small RPCs originates with , who introduce the idea of an implicit-ACK. Similar to eRPC, the Sprite RPC system  directly uses raw datagrams and performs retransmissions only at clients. The Direct Access File System  was o...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.1145/2080.357392", "end": 169, "openalex_id": "https://openalex.org/W2106035061", "raw": "A. D. Birrell and B. J. Nelson. Implementing remote procedure calls. ACM Trans. Comput. Syst., 1984.", "source_ref_id": "a96962f42f636b9ce5cb...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 18410, 41170, 111, 163136, 98, 627, 12233, 7, 41361, 15572, 84382, 1439, 13, 91363, 19336, 62038, 1636, 678, 65508, 70, 6528, 165164, 284, 20572, 209683, 28, 1052, 49188, 1486, 5426, 105237, 4527, 49649, 2053, 25561, 136, 51339, 456, 30145,...
[ 0.08941650390625, 0.094482421875, 0.0050048828125, 0.1552734375, 0.0202484130859375, 0.124267578125, 0.299072265625, 0.1141357421875, 0.1407470703125, 0.1646728515625, 0.045501708984375, 0.158203125, 0.1021728515625, 0.2127685546875, 0.114013671875, 0.08935546875, 0.02984619140625, ...
dd41f8c0c516710e3c22f23b2d652deeabddbbd6
subsection
48
53
Co-design.
There is a rapidly-growing list of projects that co-design distributed systems with the network. This includes key-value stores , , , , distributed databases and transaction processing systems , , , , state machine replication , , and graph-processing systems . We believe the availability of eRPC will motivate research...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 261, "openalex_id": "https://openalex.org/W2129554014", "raw": "C. Mitchell, Y. Geng, and J. Li. Using one-sided RDMA reads to build a fast, CPU-efficient key-value store. In Proc. USENIX Annual Technical Conference, San Jose, CA,...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 8622, 25545, 538, 8407, 14775, 5303, 77635, 552, 9, 34198, 15917, 3674, 76519, 678, 33120, 96853, 22799, 27494, 4343, 63399, 124161, 9433, 11341, 36279, 456, 182867, 41382, 63923, 1401, 18822, 185, 25820, 28, 1052, 12233, 1221, 29320, 25188, ...
[ 0.002685546875, 0.0869140625, 0.002471923828125, 0.0863037109375, 0.042205810546875, 0.08770751953125, 0.1634521484375, 0.1783447265625, 0.03814697265625, 0.256103515625, 0.24951171875, 0.09246826171875, 0.19140625, 0.1632080078125, 0.1641845703125, 0.02069091796875, 0.1246337890625,...
3b81c28fb20b0f773f0468d126c37453d2e9e6bc
subsection
49
53
Conclusion
eRPC is a fast, general-purpose RPC system that provides an attractive alternative to putting more functions in network hardware, and specialized system designs that depend on these functions. eRPC's speed comes from prioritizing common-case performance, carefully combining a wide range of old and new optimizations, an...
{ "cite_spans": [ { "arxiv_id": "", "doi": "", "end": 1087, "openalex_id": "https://openalex.org/W2520381032", "raw": "R. Ricci, E. Eide, and The CloudLab Team. Introducing CloudLab: Scientific infrastructure for advancing cloud architectures and applications. USENIX ;login:, 2014.",...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 28, 1052, 12233, 83, 10, 4271, 4, 4537, 4717, 78381, 627, 5426, 450, 87344, 225530, 30700, 118620, 1286, 32354, 7, 33120, 76407, 5361, 29367, 4331, 56566, 98, 38352, 32497, 1295, 25343, 84382, 39210, 9, 58437, 23718, 197918, 375, 592, 381...
[ 0.140625, 0.221923828125, 0.35888671875, 0.139404296875, 0.049652099609375, 0.1885986328125, 0.0189666748046875, 0.171142578125, 0.1627197265625, 0.00567626953125, 0.13427734375, 0.2406005859375, 0.0186767578125, 0.08978271484375, 0.1856689453125, 0.17919921875, 0.07025146484375, 0...
2a2228623034787d6475f1aefaa2e2b56ac4ac48
subsection
50
53
eRPC's NIC memory footprint
Primarily, four on-NIC structures contribute to eRPC's NIC memory footprint: the TX and RX queues, and their corresponding completion queues. The TX queue must allow sufficient pipelining to hide PCIe latency; we found that 64 entries are sufficient in all cases. eRPC's TX queue and TX completion queue have 64 entries ...
{ "cite_spans": [ { "arxiv_id": "", "doi": "10.5555/2616448.2616486", "end": 487, "openalex_id": "https://openalex.org/W1532546444", "raw": "A. Dragojević, D. Narayanan, O. Hodson, and M. Castro. FaRM: Fast remote memory. In Proc. 11th USENIX NSDI, Seattle, WA, Apr. 2014.", "so...
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 19012, 538, 22759, 98, 67099, 45646, 162466, 28, 1052, 12233, 6, 98323, 57616, 35662, 54047, 136, 627, 1542, 41, 21049, 4, 42518, 214, 21721, 1830, 13388, 8110, 63769, 129980, 137158, 150, 47, 1274, 112, 167752, 13, 19161, 14037, 7911, 11...
[ 0.012969970703125, 0.020050048828125, 0.154296875, 0.173583984375, 0.192626953125, 0.17529296875, 0.1519775390625, 0.0887451171875, 0.15478515625, 0.259521484375, 0.020111083984375, 0.1998291015625, 0.1466064453125, 0.1737060546875, 0.197509765625, 0.05364990234375, 0.09979248046875,...
2c8623a55fac24322a0578ac4d8cc9dc15d54cae
subsection
51
53
Handling node failures
eRPC launches a session management thread that handles sockets-based management messaging for creating and destroying sessions, and detects failure of remote nodes with timeouts. When the management thread suspects a remote node failure, each dispatch thread with sessions to the remote node acts as follows. First, it f...
{ "cite_spans": [] }
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 28, 1052, 12233, 83184, 90, 56002, 24365, 86997, 34831, 221, 27853, 77007, 115288, 9966, 105233, 163684, 104227, 96391, 137578, 148814, 110, 988, 1733, 6056, 14847, 92610, 7, 112, 12638, 58580, 257, 47, 27992, 28960, 23972, 14838, 114942, 540...
[ 0.0882568359375, 0.158447265625, 0.285400390625, 0.1524658203125, 0.044219970703125, 0.2135009765625, 0.2213134765625, 0.22265625, 0.10333251953125, 0.0506591796875, 0.1654052734375, 0.07952880859375, 0.0867919921875, 0.05474853515625, 0.08837890625, 0.131103515625, 0.1842041015625, ...
ee071cdc21669366bd5b61a17d28afe985a753d4
subsection
52
53
Rate limiting with zero-copy
Recall the request retransmission example discussed in § REF : On receiving the response for the first copy of a retransmitted request, we wish to ensure that the rate limiter does not contain a reference to the retransmitted copy. Unlike eRPC's NIC DMA queue that holds only a few tens of packets, the rate limiter trac...
{ "cite_spans": [] }
1806.00680
Datacenter RPCs can be General and Fast
[ "Anuj Kalia", "Michael Kaminsky", "David G. Andersen" ]
[ "cs.OS" ]
2,018
en
Computer Science
[ 85763, 50336, 456, 30145, 21150, 27781, 45252, 5360, 9069, 919, 217063, 57553, 5117, 43658, 4007, 32599, 63284, 34515, 17475, 56, 959, 70541, 91067, 5062, 28, 1052, 12233, 67099, 391, 8218, 41, 13388, 16401, 10846, 93396, 43824, 28560, 8877, ...
[ 0.1385498046875, 0.1883544921875, 0.050140380859375, 0.09893798828125, 0.055938720703125, 0.1697998046875, 0.0184783935546875, 0.0305938720703125, 0.0452880859375, 0.112548828125, 0.011627197265625, 0.1624755859375, 0.025970458984375, 0.1029052734375, 0.03936767578125, 0.04440307617187...
5c017091042bd3301b707a1a3a52d4036bbde4fa
abstract
0
63
Abstract
We analyze binary data, available for a relatively large number (big data) of families (or households), which are within small areas, from a population-based survey. Inference is required for the finite population proportion of individuals with a specific character for each area. To accommodate the binary data and impo...
{ "cite_spans": [] }
1806.00446
Bayesian Logistic Regression for Small Areas with Numerous Households
[ "Balgobin Nandram", "Lu Chen", "Shuting Fu", "Binod Manandhar" ]
[ "stat.ME" ]
2,018
en
Statistics
[ 1401, 7968, 53, 731, 2394, 6635, 2053, 19882, 100, 35845, 21334, 32976, 87143, 748, 197540, 28032, 19336, 58555, 43904, 77007, 110297, 360, 69988, 56065, 94418, 123875, 79165, 29458, 62816, 16128, 211196, 136, 5526, 66139, 756, 121413, 4527, ...
[ 0.0557861328125, 0.1544189453125, 0.0965576171875, 0.015045166015625, 0.23388671875, 0.199462890625, 0.2147216796875, 0.1409912109375, 0.012054443359375, 0.047210693359375, 0.041534423828125, 0.029815673828125, 0.1434326171875, 0.025604248046875, 0.1507568359375, 0.0465087890625, 0.1...
145aa0a57076d715f0bd034ac1a0475420e132a1
subsection
1
63
-5pt
12pt plus 4pt minus 2pt0pt plus 2pt minus 2pt12pt plus 4pt minus 2pt0pt plus 2pt minus 2pt12pt plus 4pt minus 2pt0pt plus 2pt minus 2pt2017] Bayesian Logistic Regression for Small Areas with Numerous Households BALGOBIN NANDRAM,  LU CHEN,  SHUTING FU AND BINOD MANANDHAR[Vol. .., Nos. 1&2firstpage Statistics and Applic...
{ "cite_spans": [] }
1806.00446
Bayesian Logistic Regression for Small Areas with Numerous Households
[ "Balgobin Nandram", "Lu Chen", "Shuting Fu", "Binod Manandhar" ]
[ "stat.ME" ]
2,018
en
Statistics
[ 427, 6328, 1001, 201, 26948, 116, 2389, 1530, 3259, 9631, 90, 3378, 13146, 48242, 853, 148448, 105508, 79200, 7, 678, 93511, 10821, 13038, 16200, 85674, 19930, 119603, 225192, 594, 30684, 215180, 159, 27204, 68151, 88082, 335, 53878, 397, 2...
[ 0.20166015625, 0.178955078125, 0.184326171875, 0.1856689453125, 0.207275390625, 0.119384765625, 0.064453125, 0.1385498046875, 0.1307373046875, 0.1715087890625, 0.12109375, 0.116455078125, 0.19482421875, 0.1297607421875, 0.119384765625, 0.27001953125, 0.201904296875, 0.181884765625,...
300f3590e098647d88eab57c8b09b04d20496b0b
subsection
2
63
Introduction
In the second Nepal Living standards Survey (NLSS II), there are data from households. One question of interest is health status (good versus poor health), a binary variable, and there are several covariates that can explain the binary outcomes. Our interest is to provide smoothed estimates of the household proportions...
{ "cite_spans": [] }
1806.00446
Bayesian Logistic Regression for Small Areas with Numerous Households
[ "Balgobin Nandram", "Lu Chen", "Shuting Fu", "Binod Manandhar" ]
[ "stat.ME" ]
2,018
en
Statistics
[ 70, 17932, 56250, 43722, 5570, 7, 181842, 39102, 15268, 1995, 247, 2685, 621, 2053, 1295, 197540, 6561, 9655, 111, 33946, 16227, 10778, 70810, 119475, 70425, 10, 2394, 6635, 77336, 4, 136, 40368, 552, 124640, 1636, 450, 831, 73342, 184345, ...
[ 0.0220489501953125, 0.0794677734375, 0.214599609375, 0.12353515625, 0.120361328125, 0.080810546875, 0.1317138671875, 0.1151123046875, 0.1776123046875, 0.12939453125, 0.022064208984375, 0.0219879150390625, 0.0219268798828125, 0.1298828125, 0.0219268798828125, 0.177490234375, 0.0120849...
bc00b2e2d12f633743a0979d17ed5e7e1a858168
subsection
3
63
Introduction
Roberts, Rao and Kumar (1987) discussed logistic regression for sample survey data (not small area estimation). Nandram and Chen (1996) show how to accelerate the Gibbs sampler for a model with latent variables introduced earlier by Albert and Chib (1993) for Bayesian probit analysis.Albert and Chib (1993) started an i...
{ "cite_spans": [] }
1806.00446
Bayesian Logistic Regression for Small Areas with Numerous Households
[ "Balgobin Nandram", "Lu Chen", "Shuting Fu", "Binod Manandhar" ]
[ "stat.ME" ]
2,018
en
Statistics
[ 106878, 72733, 136, 18385, 208000, 45252, 297, 150535, 456, 148448, 121413, 110297, 2053, 15, 10869, 19336, 16128, 25902, 1363, 353, 14519, 39, 123378, 188078, 7639, 3642, 47, 197108, 13, 70, 92260, 16145, 42, 100, 10, 3299, 678, 19161, 18,...
[ 0.18505859375, 0.1490478515625, 0.09478759765625, 0.123779296875, 0.146484375, 0.0859375, 0.0262451171875, 0.330078125, 0.148681640625, 0.29345703125, 0.2193603515625, 0.2059326171875, 0.2249755859375, 0.0262298583984375, 0.041778564453125, 0.11669921875, 0.09588623046875, 0.111206...
2ddece716e5c2980808d54a6ba77fd66d217105c
subsection
4
63
Introduction
Yet INLA has found many useful applications. See, for example, Fong, Rue and Wakefield (2010) for an application on Poisson regression, and Illian, Sørbye and Rue (2012) for a realistic application on spatial point pattern data. We note that INLA can be problematic especially for logistic and Poisson hierarchical regre...
{ "cite_spans": [] }
1806.00446
Bayesian Logistic Regression for Small Areas with Numerous Households
[ "Balgobin Nandram", "Lu Chen", "Shuting Fu", "Binod Manandhar" ]
[ "stat.ME" ]
2,018
en
Statistics
[ 63519, 5881, 8356, 1556, 14037, 5941, 80234, 86685, 6872, 4, 100, 27781, 563, 4021, 4518, 13, 136, 145302, 28394, 71189, 142, 38415, 64249, 1681, 456, 148448, 891, 38820, 44022, 1272, 75582, 10, 61207, 1771, 98, 5623, 289, 6275, 103510, 2...
[ 0.089599609375, 0.2235107421875, 0.295166015625, 0.0210723876953125, 0.09033203125, 0.09967041015625, 0.19384765625, 0.2340087890625, 0.045196533203125, 0.051177978515625, 0.0509033203125, 0.1531982421875, 0.03082275390625, 0.18115234375, 0.1409912109375, 0.0960693359375, 0.051208496...
bb70ef6d7008bf8fb0cc6c328b85884fba5a0fd6
subsection
5
63
Approximate Theory and Method
The method we developed here for many small areas can be applied to any generalized linear model in the same manner. Of course, the specific models will be different. For example, for the model for Poisson regression is different from the model for logistic regression. However, note that for logistic regression model, ...
{ "cite_spans": [] }
1806.00446
Bayesian Logistic Regression for Small Areas with Numerous Households
[ "Balgobin Nandram", "Lu Chen", "Shuting Fu", "Binod Manandhar" ]
[ "stat.ME" ]
2,018
en
Statistics
[ 55300, 642, 126809, 3688, 5941, 19336, 58555, 831, 186, 190659, 47, 2499, 4537, 192617, 3299, 23, 70, 5701, 144996, 6619, 4, 29458, 115774, 1221, 12921, 1326, 64249, 1681, 456, 148448, 83, 1295, 150535, 33306, 20537, 100, 25072, 17366, 1915...
[ 0.1683349609375, 0.031524658203125, 0.1114501953125, 0.09759521484375, 0.09356689453125, 0.1156005859375, 0.138916015625, 0.0755615234375, 0.00274658203125, 0.150634765625, 0.001129150390625, 0.06573486328125, 0.1138916015625, 0.189453125, 0.231689453125, 0.00189208984375, 0.00204467...
5019bf353a773b1fa15ba1523e932cf8587f6f75
subsection
6
63
Approximate Theory and Method
Let y_{ij} and {x_\crcr \vbox to.2ex{\hbox{$x\tilde{}$}\vss }}{ij} = (1,x_{ij1},\dots ,x_{ijp-1} )^T, denote the responses and the p vector of covariates with an intercept (x_{ij0}=1).A standard hierarchical Bayesian logistic regression model isy_{ij}|{\beta ,\crcr \vbox to.2ex{\hbox{$\beta \tilde{}$}\vss }}\nu _i \sta...
{ "cite_spans": [] }
1806.00446
Bayesian Logistic Regression for Small Areas with Numerous Households
[ "Balgobin Nandram", "Lu Chen", "Shuting Fu", "Binod Manandhar" ]
[ "stat.ME" ]
2,018
en
Statistics
[ 10842, 113, 454, 13786, 136, 425, 23150, 6, 334, 11728, 47, 304, 3355, 41872, 3675, 112, 24854, 8152, 4369, 18043, 7, 47391, 2203, 2858, 418, 4, 15464, 46802, 5759, 1388, 618, 8, 48345, 70, 57553, 915, 173, 18770, 111, 552, 124640, 16...
[ 0.097412109375, 0.1153564453125, 0.020965576171875, 0.2391357421875, 0.1121826171875, 0.08575439453125, 0.1319580078125, 0.021087646484375, 0.0595703125, 0.1966552734375, 0.08929443359375, 0.1409912109375, 0.144287109375, 0.0213165283203125, 0.2139892578125, 0.26806640625, 0.02142333...
1f816493b8553700937b5be226be8d5df2f354b5
subsection
7
63
Approximate Theory and Method
Omitting the intercept term from the covariate x_x\tilde{}ij, we have y_{ij}|\mu _i,{\beta _\crcr \vbox to.2ex{\hbox{$\beta \tilde{}$}\vss }}{(0)} \stackrel{ind}{\sim } \mbox{Bernoulli} \left\lbrace \frac{e^{{x_\crcr \vbox to.2ex{\hbox{$x\tilde{}$}\vss }}{ij}^{\prime }{\beta _\crcr \vbox to.2ex{\hbox{$\beta \tilde{}$}...
{ "cite_spans": [] }
1806.00446
Bayesian Logistic Regression for Small Areas with Numerous Households
[ "Balgobin Nandram", "Lu Chen", "Shuting Fu", "Binod Manandhar" ]
[ "stat.ME" ]
2,018
en
Statistics
[ 2383, 14, 19514, 1940, 19462, 13579, 1295, 552, 124640, 67, 1022, 454, 425, 3675, 112, 13786, 642, 765, 113, 58745, 561, 101, 24854, 59865, 23150, 334, 11728, 47, 5, 304, 3355, 127, 41872, 18043, 177609, 590, 2594, 7962, 7250, 5072, 39,...
[ 0.1209716796875, 0.157470703125, 0.08160400390625, 0.1478271484375, 0.2442626953125, 0.1788330078125, 0.019012451171875, 0.1390380859375, 0.256103515625, 0.075927734375, 0.101318359375, 0.15625, 0.113037109375, 0.1900634765625, 0.226806640625, 0.22607421875, 0.05950927734375, 0.066...
a39cbafd7330498fd249ee6943298ae90cd9bea9
subsection
8
63
Approximate Theory and Method
Because we are not linking the census to the NLSS, we do not have the covariates and the number of members in each nonsampled households, both being obtained using a Bayesian bootstrap (Rubin 1981) of the original samples.To develop the approximate methodology, we will work with the no-intercept model. Then, using Baye...
{ "cite_spans": [] }
1806.00446
Bayesian Logistic Regression for Small Areas with Numerous Households
[ "Balgobin Nandram", "Lu Chen", "Shuting Fu", "Binod Manandhar" ]
[ "stat.ME" ]
2,018
en
Statistics
[ 88949, 642, 621, 959, 3126, 149039, 7, 47, 39102, 15268, 54, 765, 552, 124640, 1636, 14012, 43032, 12638, 351, 25133, 6259, 197540, 15044, 113054, 17368, 9631, 90, 3378, 49935, 107617, 26951, 7568, 26771, 7311, 121413, 85493, 35707, 5134, 5...
[ 0.0755615234375, 0.02069091796875, 0.04266357421875, 0.1639404296875, 0.1514892578125, 0.1875, 0.1680908203125, 0.05413818359375, 0.1689453125, 0.2437744140625, 0.03436279296875, 0.0653076171875, 0.041778564453125, 0.18310546875, 0.048126220703125, 0.0035400390625, 0.1585693359375, ...