Split (1)

train · 617k rows

text stringlengths 2 132k	source dict
which is trusted by everybody. This failure detector was shown to be equivalent to 3S in the proof that it is the weakest to solve consensus [Chu 1998]. Aguilera et al. [2000b] presented a failure detector called heartbeat which is use-ful in designing protocols which are quiescent , i.e., which eventually stop sending...	{ "page_id": null, "source": 7326, "title": "from dpo" }
a stable period was [[ x seconds, ]] a time which is usually sufficient for an algorithm to terminate, e.g., a transaction to commit. Therefore, an arguably assumption is that the system is synchronous “forever” after an initial finite time of asynchrony. This is captured in what became known as the assumption of parti...	{ "page_id": null, "source": 7326, "title": "from dpo" }
layers and leave the application only with those issues which it needs to care The Failure Detector Abstraction · 13 about: reasoning about failures. System engineers only need to agree on the inter-face of the particular failure detector in question, and two groups can independently go about designing solutions: one g...	{ "page_id": null, "source": 7326, "title": "from dpo" }
non-blocking atomic commit using ? Pand a consensus abstraction [Guer-raoui 2002]. Finally, we argue here that failure detectors also remedy the problems of asym-metric or differing timeouts within an application which was discussed at the end of the previous section. While failure detectors do not offer timing informa...	{ "page_id": null, "source": 7326, "title": "from dpo" }
interface which makes it easier to design, build and analyse fault-tolerant distributed systems. Secondly, implementation of the failure detection functionality can be done in a centralized, re-useable fashion which enables solutions which are more efficient compared to situations in which every application performs fa...	{ "page_id": null, "source": 7326, "title": "from dpo" }
failures. We do not consider Byzantine failures: a process either correctly executes the algorithm assigned to it, or crashes and stops forever executing any action. A process that does not crash in a given execution is called correct . A process that is not correct is called faulty . A failure pattern F , a proper sub...	{ "page_id": null, "source": 7326, "title": "from dpo" }
only before its k-th element, i.e., ∀k′ ≥ k, pik′ 6 = p. Failure detectors. A failure detector D with range RD is a function that maps each failure pattern F to an F -complete set of failure detector histories with range RD . D(F ) is thus the set of failure detector histories permitted by D for failure pattern F . Not...	{ "page_id": null, "source": 7326, "title": "from dpo" }
(∀k′ ≥ k : p 6 = pik′ )) ) —The eventually perfect failure detector 3P [Chandra and Toueg 1996] also outputs a set of suspected processes at each process. But the guarantees provided by 3P are weaker than those of P. There is a time after which 3P outputs the set of all faulty processes at every non-faulty process. Mor...	{ "page_id": null, "source": 7326, "title": "from dpo" }
\ F ). —The failure signal failure detector FS [Delporte-Gallet et al. 2004], originally called anonymously perfect failure detector in [Guerraoui 2002] outputs green or red at each process. As long as there are no failures, FS outputs green at every process; after a failure occurs, and only if it does, FS must eventua...	{ "page_id": null, "source": 7326, "title": "from dpo" }
the set of messages currently in the message buffer. Initially, the message buffer is empty. A step ( p, m, d ) of an algorithm A is uniquely determined by the identity of the process p that takes the step, the message m received by p during the step ( m might be the null message λ), and the The Failure Detector Abstra...	{ "page_id": null, "source": 7326, "title": "from dpo" }
a failure detector history H ∈ D (F ). A run of algorithm A in an environment E using a failure detector D is a tuple 〈F, I, S 〉 where F ∈ E , I is an initial configuration of A, and S = ( pi1 , m 1, d 1), (pi1 , m 1, d 1), . . . is an infinite schedule of A, applicable to I, such that ( pi1 , d 1), (pi1 , d 1), . . . ...	{ "page_id": null, "source": 7326, "title": "from dpo" }
to emulate all histories of D;it is required that all the histories it emulates be histories of D. A weakest failure detector. We say that a failure detector D is the weakest failure detector to solve a problem M in an environment E if the following conditions are satisfied: (a) D is sufficient to solve M in E, i.e., D...	{ "page_id": null, "source": 7326, "title": "from dpo" }
the identifier of a process so that eventually, the identifier of the same correct process is output permanently at all correct processes. The basic idea underlying TD→ Ω is to have each process locally simulate the overall distributed system in which the processes execute several runs of A that could have happened in ...	{ "page_id": null, "source": 7326, "title": "from dpo" }
between them. All this information is pieced together in a single data structure, a directed acyclic graph (DAG) Gp.Informally, every vertex [ q, d, k ] of Gp is a failure detector value “seen” by q in its k-th query of its failure detector module. An edge ([ q, d, k ], [q′, d ′, k ′]) can be interpreted as “ q saw fai...	{ "page_id": null, "source": 7326, "title": "from dpo" }
. be any path in G. Then H = ( q1, d 1) → (q2, d 2) → . . . is a partial failure detector history in D(F ). (5) For all correct processes p and q, and for every vertex v of Gp, there is a d ∈ R D and a k ∈ N such that eventually ( v, [q, d, k ]) is an edge of Gq .Note that properties (1)–(5) imply that, for every corre...	{ "page_id": null, "source": 7326, "title": "from dpo" }
Gp and I. The set of vertices of Υ Ip is the set of all schedules S that are applicable to I and compatible with paths in Gp. The root of Υ Ip is the empty schedule S⊥. There is an edge from S to S′ if and only if S′ = S · e for a step e; the edge is labeled e. Thus, every vertex S of Υ Ip is associated with a sequence...	{ "page_id": null, "source": 7326, "title": "from dpo" }
[p1, d 3, k 3]. Every path through the DAG and an initial configuration I induce at least one schedule in the simulation tree. Hence, the simulation tree has at least three leaves: (p1, λ, d 1) ( p2, m 2, d 2) ( p1, m 3, d 3), ( p2, λ, d 2) ( p1, m ′ > 3 , d 3), and (p1, λ, d 3). Recall that λ is the empty message: sin...	{ "page_id": null, "source": 7326, "title": "from dpo" }
are not sufficiently long yet. But this is just a matter of time: if p is correct, then every vertex of p’s simulation forest will eventually have an extension in which correct processes appear sufficiently often for p to take a decision. A vertex S of Υ ip is 0 -valent if it has exactly one tag {0} (only 0 can be deci...	{ "page_id": null, "source": 7326, "title": "from dpo" }
{0, 1, . . . , n }, and S be any vertex of Υ ip. Then: (i) There exists a non-empty V ⊆ { 0, 1} such that eventually the valence of S is permanently V . (We say that the valence of S stabilizes on V at p.) (ii) If the valence of S stabilizes on V at p, then for every correct process q, even-tually S is a vertex of Υ iq...	{ "page_id": null, "source": 7326, "title": "from dpo" }
and Υ i is either 1-valent or bivalent. Since tagged simulation forests computed at the correct processes tend to the same infinite tagged simulation forest, eventually, all correct processes compute the same smallest critical index i of the same type (univalent or bivalent). Now we have two cases to consider for the s...	{ "page_id": null, "source": 7326, "title": "from dpo" }
fork (case (a) in Figure 6) consists of a bivalent vertex S from which two different steps by the same process q, consuming the same message m, are possible which lead, on the one hand, to a 0-valent vertex S0 and, on the other hand, to a 1-valent vertex S1.A hook (case (b) in Figure 6) consists of a bivalent vertex S,...	{ "page_id": null, "source": 7326, "title": "from dpo" }
S1(Ii). Since g1 · ˜g is a path of G and ˜S is compatible with a prefix The Failure Detector Abstraction · 23 of ˜ g, it follows that S1 · ˜S is a vertex of Υ i. Hence, p also decides 0 in S1 · ˜S(Ii). But S1 is 1-valent — a contradiction. Let γ be a hook (case (b) in Figure 6). Then there is a finite schedule ˜S compa...	{ "page_id": null, "source": 7326, "title": "from dpo" }
> S′·(p, m, d ) is a bivalent vertex of Υ i〉 > then S←S′·(p, m, d ) > else exit > Fig. 7. Locating a decision gadget We show that the algorithm indeed terminates. Suppose not. Then the algorithm locates an infinite fair path through the simulation tree, i.e., a path in which all correct processes get scheduled infinite...	{ "page_id": null, "source": 7326, "title": "from dpo" }
d′, S′ · (p, m, d ′) is a vertex of Υ i. In both cases, we obtain S′ such that for some d′, S′ · (p, m, d ′) is a 1-valent vertex of Υ i.Let the path from S to S′ go through the vertices σ0 = S, σ 1, . . . , σ m−1, σ m = S′.By transitivity of G, for all k ∈ { 0, 1, . . . , m }, σk · (p, m, d ′) is a vertex of Υ i. By s...	{ "page_id": null, "source": 7326, "title": "from dpo" }
situation. Step ( p, m, d ′) brings σk−1 to a 0-valent state, and σk = σk−1 · (p′, m ′, d ′′ ) to a 1-valent state (Case 2 in Figure 8). But that is a hook! As a result, any bivalent infinite simulation tree has at least one decision gad-get. 3.2.9 The reduction algorithm. Now we are ready to complete the description o...	{ "page_id": null, "source": 7326, "title": "from dpo" }
If the forest has a bivalent critical index i and Υ ip contains a decision gadget, then p outputs the deciding process of the smallest decision gadget in Υ ip (the “smallest” can be well-defined, since the vertices of the simulation tree are countable). Eventually, the correct processes locate the same stable critical ...	{ "page_id": null, "source": 7326, "title": "from dpo" }
specific value to be stored in the register and returns a simple indication ok that the operation has been executed. The read operation takes no parameters and returns a value according to one of the following consistency criteria. A (single-writer, multi-reader) safe register ensures only that any read operation that ...	{ "page_id": null, "source": 7326, "title": "from dpo" }
algorithm that implements registers in any environment using Σ. Where the original algorithm uses waiting until a majority responds to ensure that a read operation returns the most recently written value, we can use the quorums provided by Σ to the same effect. 3.3.3 The reduction algorithm. Now we need to show that an...	{ "page_id": null, "source": 7326, "title": "from dpo" }
in which p completes A, is called a complete p-solo run of A.It is important to notice that in any run R of A in which two processes p and q both complete executing A, either p reads 1 in Xq , or q reads 1 in Xp. Intuitively, this implies that the sets of processes “involved” in the executions of A at p and q intersect...	{ "page_id": null, "source": 7326, "title": "from dpo" }
Σ -output p and Σ -output q always intersect. Indeed, assume, by contradiction, that there exist P, Q ⊂ Π such that P ∩Q = ∅,and, at some time t1, p computes Σ -output p = P and, at some time t2, q computes Σ -output q = Q.By the algorithm of Figure 10, A has a complete p-solo run Rp = 〈F, I, S p〉28 · Freiling, Guerrao...	{ "page_id": null, "source": 7326, "title": "from dpo" }
On the other hand, consensus can be used to implement atomic registers in any environment [Lamport 1978; Schneider 1990], and thus to extract Σ. Combined with the fact that Ω is necessary to solve consensus in any environment [Chandra et al. 1996] (see Section 3.2), this implies that (Ω , Σ) is necessary to solve conse...	{ "page_id": null, "source": 7326, "title": "from dpo" }
or FS detects a failure by outputting red . If the votes of all processes are received and are yes , then p sets the myproposal variable to The Failure Detector Abstraction · 29 > non-blocking atomic commit (v): {vis yes or no } > 1send vto all > 2wait until [(for each process qin Π, received q’s vote) or FS p=red ] > ...	{ "page_id": null, "source": 7326, "title": "from dpo" }
1 is decided in the consensus algorithm, then p returns commit . If 0 is decided, then p returns abort .The Agreement property of NBAC follows from the Agreement property of con-sensus and the fact that the output of Ψ switches uniformly from ⊥ to (Ω , Σ) or FS at all processes. If there are no failures, then eventuall...	{ "page_id": null, "source": 7326, "title": "from dpo" }
there views of proposed values (the (Ω , Σ) part of failure detector Ψ). Let E be any environment. Let D be any failure detector that solves NBAC in E, and let A be any algorithm that solves NBAC in E using D.There is a straightforward reduction algorithm that transforms D into FS [Charron-Bost and Toueg 2001; Guerraou...	{ "page_id": null, "source": 7326, "title": "from dpo" }
[Chandra and Toueg 1996] defines a failure detector to be a mapping from a failure pattern F to some output range H. The failure pattern F specifies which processes fail at what time. So anything that can be defined as a function of failures can be formally called a failure detector. Not everything that looks like a fa...	{ "page_id": null, "source": 7326, "title": "from dpo" }
(i.e. having a bound δ but not having a bound ∆, see Section 2.1.5). So at least a perfect failure detector can be regarded as an abstraction of process synchrony, not of channel synchrony. 4.2 Do Failure Detectors make sense outside of the crash model? The “classic” failure detectors have been crash failure detectors,...	{ "page_id": null, "source": 7326, "title": "from dpo" }
of unbounded counters hinting on how often a process has been suspected. Considering the consensus problem, failure detector specifications have been ex-tended to environments where the network may partition [Guerraoui and Schiper 1996; Aguilera et al. 1999] and processes may experience send and receive omissions [Dole...	{ "page_id": null, "source": 7326, "title": "from dpo" }
application level messages. The closest we can get are so-called “muteness” detectors [Doudou et al. 1999; Doudou et al. 2002]. But this approach usually assumes that processes send certain messages continuously and relay messages to all others if they receive them for the first time. Kihlstrom et al. distinguishes be...	{ "page_id": null, "source": 7326, "title": "from dpo" }
the relationship between fairness and randomization. He showed that to solve crash-tolerant consensus it is sufficient to postulate a certain kind of fairness called hyperfairness . Briefly spoken, hyperfairness looks at situa-tions in which certain resources are needed (e.g., for a process to terminate) but which can ...	{ "page_id": null, "source": 7326, "title": "from dpo" }
(eventually) timely. As discussed above, perfect failure detectors allow building a system which is very close (but not equivalent) to a fully synchronous one [Charron-Bost et al. 2000]. It has still been argued that the use of failure detectors also offers the potential of building real-time applications by using the ...	{ "page_id": null, "source": 7326, "title": "from dpo" }
Symposium on Distributed Computing . LNCS. Aguilera , Delporte-Gallet , Fauconnier , and Toueg . 2003. On implementing omega with weak reliability and synchrony assumptions. In PODC: 22th ACM SIGACT-SIGOPS Sympo-sium on Principles of Distributed Computing . Aguilera, M. and Toueg, S. 1998. Failure detection and randomi...	{ "page_id": null, "source": 7326, "title": "from dpo" }
124–142. Attiya, H. and Welch, J. L. 2004. Distributed Computing: Fundamentals, Simulations and Advanced Topics (2nd edition) . Wiley. Barborak, M. , Dahbura, A. , and Malek, M. 1993. The consensus problem in fault-tolerant computing. ACM Computing Surveys 25, 2 (June), 171–220. Beauquier, J. and Kekkonen-Moneta, S. 19...	{ "page_id": null, "source": 7326, "title": "from dpo" }
67 , 289–293. Cristian, F. and Fetzer, C. 1999. The timed asynchronous distributed system model. IEEE Transactions on Parallel and Distributed Systems 10, 6 (June). Delporte-Gallet, C. , Fauconnier, G. , and Freiling, F. C. 2005. Revisiting failure detection and consensus in omission failure environments. In Theoretica...	{ "page_id": null, "source": 7326, "title": "from dpo" }
and Stockmeyer, L. 1988. Consensus in the presence of partial syn-chrony. Journal of the ACM 35, 2 (Apr.), 288–323. Eisler, J. , Hadzilacos, V. , and Toueg, S. 2004. The quorum failure detector and its relation to consensus and registers. Unpublished note. Fischer, M. J. , Lynch, N. A. , and Paterson, M. S. 1985. Impos...	{ "page_id": null, "source": 7326, "title": "from dpo" }
in asynchronous systems with failure detectors. Distributed Computing 15, 1, 17–25. Guerraoui, R. and Schiper, A. 1996. “Gamma-accurate” failure detectors. In Distributed Algorithms, 10th International Workshop, WDAG ’96 , ¨O. Babaoglu and K. Marzullo, Eds. Lecture Notes in Computer Science, vol. 1151. Springer-Verlag,...	{ "page_id": null, "source": 7326, "title": "from dpo" }
= Implementing Reliable Distributed Real-Time Systems with the Theta-Model, O. . . O. . . b. . I. O. . . y. . . O. . . O. . . O. . . O. . . O. . . m. . d. O. . . O. . . O. . . O. . . Kihlstrom, K. P. , Moser, L. E. , and Melliar-Smith, P. M. 2003. Byzantine fault detectors for solving consensus. The Computer Journal 46...	{ "page_id": null, "source": 7326, "title": "from dpo" }
Lausanne, Switzerland. Aug. Paxson, V. and Adams, A. 2002. Experiences with NIMI. In Proceedings of the 2002 Symposium on Applications and the Internet . Pedone, F. and Schiper, A. 1999. Generic broadcast. In Proceedings of the 13th International Symposium on Distributed Computing (DISC’99) . Powell, D. 1992. Failure m...	{ "page_id": null, "source": 7326, "title": "from dpo" }
and hyperfairness in distributed computing. In Proceedings of the 19th International Symposium on Distributed Computing, DISC 2005 . Number 3724 in Lecture Notes in Computer Science. Springer-Verlag, 33–47.	{ "page_id": null, "source": 7326, "title": "from dpo" }
Title: URL Source: Markdown Content: # Rachid Guerraoui, Lu´ ıs Rodrigues # Introduction to Distributed Algorithms # (Preliminary Draft) ## November 22, 2004 # Springer-Verlag ## Berlin Heidelberg NewYork London Paris Tokyo Hong Kong Barcelona Budapest To whom it might concern. DRAFT V (22/11/2004) Preface This manu...	{ "page_id": null, "source": 7326, "title": "from dpo" }
distributed programming can be significantly simplified if the difficulty of robust cooperation is encapsulated within specific abstrac-tions . By encapsulating all the tricky algorithmic issues, such distributed programming abstractions bridge the gap between network communication layers, usually frugal in terms of re...	{ "page_id": null, "source": 7326, "title": "from dpo" }
has reached a certain level of maturity where details, for instance about the network and various kinds of failures, can be abstracted away when reasoning about the distributed algorithms. Elementary notions of algorithms, first order logics, program-ming languages, networking, and operating systems might be helpful, b...	{ "page_id": null, "source": 7326, "title": "from dpo" }
used in the manuscript to describe specifications and algorithms. • In Chapter 2 we present different kinds of assumptions that we will be mak-ing about the underlying distributed environment, i.e., we present different distributed system models. Basically, we describe the basic abstractions on which more sophisticated...	{ "page_id": null, "source": 7326, "title": "from dpo" }
read-write shared memory objects. DRAFT IX (22/11/2004) Preface Preface • In Chapter 7 we gather what we call coordination abstractions. These in-clude leader election, terminating reliable broadcast, non-blocking atomic commit and group membership. The distributed algorithms we will study differ naturally according ...	{ "page_id": null, "source": 7326, "title": "from dpo" }
directly or indirectly inspired the material of this manuscript. We did our best to reference their work throughout the text. Most chapters end with a historical note. This intends to provide hints for further readings, to trace the history of the concepts presented in the chapters, as well as to give credits to those ...	{ "page_id": null, "source": 7326, "title": "from dpo" }
colleagues who were kind enough to read and comment earlier drafts of this book. These include Lorenzo Alvisi, Roberto Baldoni, Carole Delporte, Hugues Fauconnier, Pas-cal Felber, Felix Gaertner, Anne-Marie Kermarrec, Fernando Pedone, Michel Raynal, and Marten Van Steen. Rachid Guerraoui and Lu´ ıs Rodrigues DRAFT XI...	{ "page_id": null, "source": 7326, "title": "from dpo" }
81.4.1 Composition Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81.4.2 Programming Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 1.4.3 Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 1.4.4 Classes of Algorithms . . . . . . . . . ...	{ "page_id": null, "source": 7326, "title": "from dpo" }
. . . . . . . . . . . . . . . . . 19 2. Basic Abstractions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 2.1 Distributed Computation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 2.1.1 Processes and Messages . . . . . . . . . . . . . . . . . . . . . . . . . . . ...	{ "page_id": null, "source": 7326, "title": "from dpo" }
. . . . . . . . . . . . . 28 2.2.3 Crashes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 2.2.4 Recoveries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 2.3 Abstracting Communication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 2.3...	{ "page_id": null, "source": 7326, "title": "from dpo" }
. . . . . . . . . . . . . . . . . . . . . . . . 37 2.4 Timing Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 2.4.1 Asynchronous System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 Contents Contents 2.4.2 Synchronous System . . . . . . . . . . . . . . . . . . . . ....	{ "page_id": null, "source": 7326, "title": "from dpo" }
. 45 2.5.4 Eventual Leader Election . . . . . . . . . . . . . . . . . . . . . . . . . . 47 2.6 Distributed System Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 2.6.1 Combining Abstractions . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 2.6.2 Measuring Performance . . . . . . . . . . . ...	{ "page_id": null, "source": 7326, "title": "from dpo" }
. . . . . . . . . . . . . . . . . . . . . . . . 55 Perfect Failure Detector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . ...	{ "page_id": null, "source": 7326, "title": "from dpo" }
. . . 61 3.1.1 Client-Server Computing . . . . . . . . . . . . . . . . . . . . . . . . . . 61 3.1.2 Multi-Participant Systems . . . . . . . . . . . . . . . . . . . . . . . . . 62 3.2 Best-Effort Broadcast . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 3.2.1 Specification . . . . . . . . . . . ...	{ "page_id": null, "source": 7326, "title": "from dpo" }
3.4.1 Specification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 3.4.2 Fail-Stop Algorithm: All-Ack URB . . . . . . . . . . . . . . . . . 69 3.4.3 Fail-Silent Algorithm: Majority-Ack URB . . . . . . . . . . . 71 3.5 Stubborn Broadcast . . . . . . . . . . . . . . . . . . . . . . . . . . ....	{ "page_id": null, "source": 7326, "title": "from dpo" }
Reliable Broadcast . . . . . . . . . . . . . . . . . . . . . . . 75 3.7.1 Specification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 3.7.2 Fail-Recovery Algorithm: Logged Majority-Ack URB . . 76 DRAFT XIV (22/11/2004) Contents Contents 3.8 Randomized Broadcast . . . . . . . . . . . . ....	{ "page_id": null, "source": 7326, "title": "from dpo" }
. . . . . . . . . . . . . . . 85 Lazy Reliable Broadcast . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 All-Ack URB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 Majority-Ack URB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 P...	{ "page_id": null, "source": 7326, "title": "from dpo" }
. . . . . . . . . . . . . 97 Historical Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 4. Shared Memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ...	{ "page_id": null, "source": 7326, "title": "from dpo" }
. . . . . . . . . . . 108 4.2.2 Fail-Stop Algorithm: Read-One-Write-All Regular Reg-ister . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 4.2.3 Fail-Silent Algorithm: Majority-Voting Regular Register111 4.3 (1,N) Atomic Register . . . . . . . . . . . . . . . . . . . . . . . ....	{ "page_id": null, "source": 7326, "title": "from dpo" }
. . . . . . . . . . . . . . . . . . . . 124 4.4.2 Specification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124 4.4.3 Transformation: From (1,N) atomic to (N,N) atomic registers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 4.4.4 Fail-Stop Algorithm:...	{ "page_id": null, "source": 7326, "title": "from dpo" }
. . . . . . . . . 132 4.5.3 Fail-Recovery Algorithm: Logged Majority Voting . . . . 133 Hands-On . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 Solut...	{ "page_id": null, "source": 7326, "title": "from dpo" }
. . . . . . . . . . . . . . . 145 5.1.1 Specification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145 5.1.2 Fail-Stop Algorithm: Consensus with Flooding . . . . . . . 146 5.1.3 Fail-Stop Algorithm: Hierarchical Consensus . . . . . . . . . 149 5.2 Uniform Consensus . . . . . . . . . . . . ....	{ "page_id": null, "source": 7326, "title": "from dpo" }
. . . . . . . . . . . . . . . . . . . . . . . . 161 5.4.1 Specification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162 5.4.2 A randomized Consensus Algorithm . . . . . . . . . . . . . . . . 163 Hands-On . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ...	{ "page_id": null, "source": 7326, "title": "from dpo" }
. . . . . . . . . 184 6.2.3 Algorithm: Uniform Total Order Broadcast . . . . . . . . . . 185 6.3 Logged Total Order Broadcast . . . . . . . . . . . . . . . . . . . . . . . . . . . 188 6.3.1 Specification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189 6.3.2 Fail-Recovery Algorithm: Total O...	{ "page_id": null, "source": 7326, "title": "from dpo" }
. . . . . . . . . . . . . . . . . . . . . 199 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203 Historical Notes . . . . . . . . . . . . . . ....	{ "page_id": null, "source": 7326, "title": "from dpo" }
. . . . 214 7.3.3 Fail-Stop Algorithm: Monarchical Leader Election . . . . 215 7.4 Group Membership . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216 7.4.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216 7.4.2 Specification . . . . . . . . . . . ....	{ "page_id": null, "source": 7326, "title": "from dpo" }
Algorithm: TRB-Based View-Synchrony . . . . 221 7.6 Probabilistic Partial Membership . . . . . . . . . . . . . . . . . . . . . . . . . 224 7.6.1 Specification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224 7.6.2 Randomized Algorithm: Probabilistic Broadcast with Partial Membership . . . ....	{ "page_id": null, "source": 7326, "title": "from dpo" }
. . . . . . . . . . . . . . . . . . . . 233 DRAFT XVII (22/11/2004) 1. Introduction God does not often clap his hands. When he does, every body should dance (African Proverb) This chapter first motivates the need for distributed programming ab-stractions. Special attention is given to abstractions that capture the pr...	{ "page_id": null, "source": 7326, "title": "from dpo" }
synchronize their activities in a consistent way. In other words, the cooperation must be made robust to tolerate partial failures. This makes distributed computing quite hard, yet extremely stimulating, problem. As we will discuss in detail later in the manuscript, due to several factors such as the asynchrony of the ...	{ "page_id": null, "source": 7326, "title": "from dpo" }
even more reasonable that the server process keeps on providing information to the other client processes, even when some of them fail or got disconnected. The problems above are already difficult to deal with when distributed computing is limited to the interaction between two parties, such as in the client-server cas...	{ "page_id": null, "source": 7326, "title": "from dpo" }
to what is called a system model . In this book we will use mainly two abstractions to represent the underlying physical system: processes and links .The processes of a distributed program abstract the active entities that perform computations. A process may represent a computer, a processor within a computer, or simpl...	{ "page_id": null, "source": 7326, "title": "from dpo" }
addresses) and some common format for representing messages. They might also need to agree on some reliable way of exchanging messages (say to provide TCP-like semantics). DRAFT 3 (22/11/2004) 1.2. ABSTRACTIONS CHAPTER 1. INTRO • After exchanging some messages, the processes may be faced with sev-eral alternative pla...	{ "page_id": null, "source": 7326, "title": "from dpo" }
the emergence of dis-tributed programming abstractions. Examples of such applications are in-formation dissemination engines, multi-user cooperative systems, distributed shared spaces, cooperative editors, process control systems, and distributed databases. Information Dissemination. In distributed applications with in...	{ "page_id": null, "source": 7326, "title": "from dpo" }
require from a strock exchange infrastructure that information be disseminated in an ordered manner. The adequate communication abstraction that offers order-ing in addition to reliability is called total order broadcast . This abstraction captures the need to disseminate information, such that all participants can get...	{ "page_id": null, "source": 7326, "title": "from dpo" }
memory abstraction is typically accessed through read and write operations that the users exploit to store and exchange information. In its simplest form, a shared working space can be viewed as a virtual register or a distributed file system. To maintain a consistent view of the shared DRAFT 5 (22/11/2004) 1.2. ABST...	{ "page_id": null, "source": 7326, "title": "from dpo" }
are presumably supposed to fail independently. The service continuity is in a sense ensured despite the crash of a subset of the machines. No specific hardware is needed: fault-tolerance through replication is software-based. In fact, replication might also be used within an informa-tion system to improve the read-acce...	{ "page_id": null, "source": 7326, "title": "from dpo" }
algorithm is able to provide the desired abstraction and the algorithm that solves the problem can have a high complexity, e.g., in terms of the number of inter-process com-munication steps and messages. Therefore, depending on the system model, the network characteristics, and the required quality of service, the over...	{ "page_id": null, "source": 7326, "title": "from dpo" }
system models. Instead, different solutions can usually be proposed and none of these solutions might strictly be supe-rior to the others: each might have its own advantages and disadvantages, performing better under different network or load conditions, making differ-ent trade-offs between network traffic and message ...	{ "page_id": null, "source": 7326, "title": "from dpo" }
general pur-pose distributed algorithms in a pedagogical way. Trying to invent a dis-tributed programming language was not an option. Had we had the time to invent one and had we even been successful, at least one book would have been required to present the language. Therefore, we have opted to use pseudo-code to desc...	{ "page_id": null, "source": 7326, "title": "from dpo" }
to satisfy some common properties. Software Stacks. Components can be composed to build software stacks, at each process: each component represents a specific layer in the stack. The application layer is on the top of the stack whereas the networking layer is at the bottom. The layers of the distributed programming abs...	{ "page_id": null, "source": 7326, "title": "from dpo" }
handling of an event is terminated, the process keeps on checking if any other event is triggered. The code of each component looks like this: upon event 〈 Event1 , att 11, att 21, . . . 〉 do something // send some event trigger 〈 Event2 , att 12,att 22, . . . 〉; upon event 〈 Event3 , att 13, att 23, . . . 〉 do somethi...	{ "page_id": null, "source": 7326, "title": "from dpo" }
to deliver information to another component. Considering the broadcast example above, at every process that is a destination of the message, the component in charge of implementing the actual broadcast primitive will typically perform some processing to ensure the corresponding reliability guarantee, and then use an in...	{ "page_id": null, "source": 7326, "title": "from dpo" }
Confirmation: 〈lprOk , rqid 〉: Used to confirm that the printing request with identifier rqid succeeded. > Module 1.1 Interface of a printing module. > Algorithm 1.1 Printing service. > Implements: > Print (lpr). > upon event 〈lprPrint , rqid, string 〉do print string; > trigger 〈lprOk , rqid 〉; 1.4.3 Modules Not surpri...	{ "page_id": null, "source": 7326, "title": "from dpo" }
1. INTRO 1.4. SOFTWARE COMPONENTS > Module: Name: BoundedPrint (blpr). > Events: Request: 〈blprPrint , rqid, string 〉: Request a string to be printed. The token rqid is an identifier of the request. > Confirmation: 〈blprStatus , rqid, status 〉: Used to return the outcome of the printing request: Ok or Nok. > Indication...	{ "page_id": null, "source": 7326, "title": "from dpo" }
of algorithmic solutions to implement our distributed programming abstrac-tions, namely: fail-stop algorithms, designed under the assumption that pro-cesses can fail by crashing but the crashes can be reliably detected by all the other processes; fail-silent algorithms where process crashes can never be reliably detect...	{ "page_id": null, "source": 7326, "title": "from dpo" }
Hands-On ## Hands-On We have implemented several of the algorithms that we will be presenting in the book. By using these implementations, the reader has the opportunity to run and experiment the algorithms in a real setting, look at the code, make changes and improvements to the given code and, eventually, take it as ...	{ "page_id": null, "source": 7326, "title": "from dpo" }
This is illustrated in Listing 1.1. Listing 1.1. Events for the Print module. > class PrintRequestEvent extends Event { > int rid ; String data; > void setId ( int rid ); > void setStrint ( String s ); > int getId (); String getString (); > } > class PrintConfirmEvent extends Event { > int rid ; > void setId ( int rid ...	{ "page_id": null, "source": 7326, "title": "from dpo" }
layer) { super (layer); } public void handle(Event event) { if (event instanceof ChannelInit) handleChannelInit((ChannelInit)event); else if (event instanceof PrintRequestEvent) { handlePrintRequest ((PrintRequestEvent)event); }} private void handleChannelInit(ChannelInit init) { try { init . go (); } catch (AppiaEvent...	{ "page_id": null, "source": 7326, "title": "from dpo" }
rid ); > void setStatus (Status s ); > int getId (); > int getStatus (); > } We proceed to define the BoundedPrintLayer , as depicted in Listing 1.5. Since the BoundedPrint module uses the services of the basic Print module, it requires the PrintConfirmEvent produced by that module. DRAFT 17 (22/11/2004) Hands-On CHA...	{ "page_id": null, "source": 7326, "title": "from dpo" }
{ e. printStackTrace(); }}} else { PrintStatusEvent status = new PrintStatusEvent (); status .setChannel (request.getChannel()); status .setSource ( this ); status . setDir(Direction. UP); status . setId ( request.getId ()); status . setStatus ( Status.NOK); try { status . init (); status .go (); } catch (AppiaEventExc...	{ "page_id": null, "source": 7326, "title": "from dpo" }
required to perform create a channel is also depicted in Listing 1.7. Listing 1.7. Creating a PrintChannel. public class Example { public static void main(String[] args) { /∗ Create layers and put them on a array ∗/ Layer [] qos = {new PrintLayer(), new BoundedPrintLayer(), new PrintApplicationLayer() }; /∗ Create a Qo...	{ "page_id": null, "source": 7326, "title": "from dpo" }
this book we are interested in studying abstractions and algorithms that are relevant for a wide range of distributed environments. In order to achieve this goal we need to capture the fundamental characteristics of various distributed systems in some basic abstractions, and on top of which we can later define other mo...	{ "page_id": null, "source": 7326, "title": "from dpo" }

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.