Gossip, Membership, and Epidemic Systems

Classifications / Distributed Systems / Gossip, Membership, and Epidemic Systems

TRACK

Probabilistic dissemination, failure evidence, membership lifecycles, resilient overlays, epidemic repair, and production design for large changing clusters.

How can a changing cluster spread membership and repair information without requiring one coordinator or immediate global agreement?

16 lessons

Continues to Consensus and Coordination

LESSONS

1. 001
   Why Local Exchange Can Inform a Whole Cluster
   [TODO]
2. 002
   How SWIM Separates Probing from Dissemination
   [TODO]
3. 003
   How HyParView Keeps a Partial Overlay Connected
   [TODO]
4. 004
   How Overlay Shape Changes Gossip Behavior
   [TODO]
5. 005
   How Phi Turns Delay into Graded Suspicion
   [TODO]
6. 006
   Why Failure Evidence and Dissemination Are Different Jobs
   [TODO]
7. 007
   Review: Membership Lifecycles, Freshness, and Failure Evidence REVIEW
   [TODO]
8. 008
   How Lifeguard Hardens SWIM Under Local Degradation
   [TODO]
9. 009
   How Plumtree Combines Fast Broadcast with Epidemic Repair
   [TODO]
10. 010
    How Anti-Entropy Finds Missing State Efficiently
    [TODO]
11. 011
    When Gossip Can Converge Without Coordination
    [TODO]
12. 012
    How Causal Metadata Prevents Stale Updates from Winning
    [TODO]
13. 013
    Tune Gossip Without Moving the Bottleneck
    [TODO]
14. 014
    Where Gossip's Trust Model Breaks
    [TODO]
15. 015
    Review: Read Production Gossip by Its Constraints REVIEW
    [TODO]
16. 016
    Capstone: Test and Debug an Epidemic Membership System CAPSTONE
    [TODO]

← Back to Distributed Systems