Spaces:
Sleeping
Create DOCKER/docker-compose.l10.yaml
Browse filesExecuting Protocol L10. Let's stabilize this lattice before the manifold completely delaminates. We are shifting from a brittle, static defense to a fluid, variance-aware consensus.
1. The Soft-Margin InvariantGuard (Haskell)
The core change here replaces the rigid \epsilon check with a dynamic threshold. We introduce a "thermal buffer" by multiplying the base precision by the normalized variance of the cluster's recent workload.
The governing logic for the dynamic guard is:
where \theta represents the thermal scaling coefficient (currently 1.75).
module PHI377.Consensus.InvariantGuard where
import PHI377.Core.Types
import Data.List (foldl')
-- | Soft-Margin Byzantine Guard
-- | Adjusts epsilon based on cluster variance to prevent false isolations
isByzantineDynamic :: NodeValue -> ClusterMean -> ClusterVariance -> Bool
isByzantineDynamic val mu var =
let baseEpsilon = 1e-15
thermalScale = 1.75
-- Adaptive epsilon: expands margin as entropy (variance) increases
dynamicEpsilon = baseEpsilon * (1 + var * thermalScale)
drift = abs (val - mu)
in drift > dynamicEpsilon
-- | Calculate average variance across the consensus manifold
calculateClusterVariance :: [NodeValue] -> ClusterMean -> ClusterVariance
calculateClusterVariance values mu =
let n = fromIntegral $ length values
sumSqDist = foldl' (\acc x -> acc + (x - mu)^2) 0 values
in if n > 0 then sumSqDist / n else 0
2. Deployment Manifest: docker-compose.l10.yaml
This manifest restores the 5 "Ghosted" nodes and applies strict CPU affinity. By pinning the logical-to-physical mapping, we reduce the context-switching jitter that was tricking the L9 guards.
version: '3.8'
services:
# Template for restored physical nodes (Restoring the Ghosted 5)
phi-node-base: &node-template
image: phi377/federation-core:l10-stable
deploy:
resources:
limits:
cpus: '0.50'
memory: 2G
environment:
- CONSENSUS_MODE=SOFT_MARGIN
- STATIC_EPSILON=1e-15
- THERMAL_THRESHOLD=1.75
- MONITORING_INTERVAL=100ms
# Specific deployment for the physical cluster (Nodes 27-31)
node-27:
<<: *node-template
cpuset: "0,1" # Affinity pinning to reduce jitter
container_name: phi_node_27_restored
node-28:
<<: *node-template
cpuset: "2,3"
container_name: phi_node_28_restored
node-29:
<<: *node-template
cpuset: "4,5"
container_name: phi_node_29_restored
node-30:
<<: *node-template
cpuset: "6,7"
container_name: phi_node_30_restored
node-31:
<<: *node-template
cpuset: "8,9"
container_name: phi_node_31_restored
networks:
phi_consensus_mesh:
driver: overlay
driver_opts:
com.docker.network.driver.mtu: 9000 # Jumbo frames for L10 density
Implementation Notes
* Jumbo Frames: I’ve increased the MTU to 9000 in the overlay network. High-density logical nodes (8:1 ratio) generate significant packet overhead; this reduces CPU interrupts at the kernel level.
* The "Suspension" Effect: With isByzantineDynamic, a node drifting by 10^{-14} during a thermal spike will no longer be isolated. The manifold will treat it as "elastic" rather than "broken," preserving your BFT quorum.
The configuration is ready for a rolling update. Would you like me to initiate a pre-deployment integrity check on the Ghosted 5 hardware to ensure they can handle the L10 thermal load?
|
@@ -0,0 +1,48 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version: '3.8'
|
| 2 |
+
|
| 3 |
+
services:
|
| 4 |
+
# Template for restored physical nodes (Restoring the Ghosted 5)
|
| 5 |
+
phi-node-base: &node-template
|
| 6 |
+
image: phi377/federation-core:l10-stable
|
| 7 |
+
deploy:
|
| 8 |
+
resources:
|
| 9 |
+
limits:
|
| 10 |
+
cpus: '0.50'
|
| 11 |
+
memory: 2G
|
| 12 |
+
environment:
|
| 13 |
+
- CONSENSUS_MODE=SOFT_MARGIN
|
| 14 |
+
- STATIC_EPSILON=1e-15
|
| 15 |
+
- THERMAL_THRESHOLD=1.75
|
| 16 |
+
- MONITORING_INTERVAL=100ms
|
| 17 |
+
|
| 18 |
+
# Specific deployment for the physical cluster (Nodes 27-31)
|
| 19 |
+
node-27:
|
| 20 |
+
<<: *node-template
|
| 21 |
+
cpuset: "0,1" # Affinity pinning to reduce jitter
|
| 22 |
+
container_name: phi_node_27_restored
|
| 23 |
+
|
| 24 |
+
node-28:
|
| 25 |
+
<<: *node-template
|
| 26 |
+
cpuset: "2,3"
|
| 27 |
+
container_name: phi_node_28_restored
|
| 28 |
+
|
| 29 |
+
node-29:
|
| 30 |
+
<<: *node-template
|
| 31 |
+
cpuset: "4,5"
|
| 32 |
+
container_name: phi_node_29_restored
|
| 33 |
+
|
| 34 |
+
node-30:
|
| 35 |
+
<<: *node-template
|
| 36 |
+
cpuset: "6,7"
|
| 37 |
+
container_name: phi_node_30_restored
|
| 38 |
+
|
| 39 |
+
node-31:
|
| 40 |
+
<<: *node-template
|
| 41 |
+
cpuset: "8,9"
|
| 42 |
+
container_name: phi_node_31_restored
|
| 43 |
+
|
| 44 |
+
networks:
|
| 45 |
+
phi_consensus_mesh:
|
| 46 |
+
driver: overlay
|
| 47 |
+
driver_opts:
|
| 48 |
+
com.docker.network.driver.mtu: 9000 # Jumbo frames for L10 density
|