Distributed computing: from practice to theory

RubyBarcamp, Altoros, Minsk 2015/01/15

Transcript

1. Distributed Computing: from practice to theory Alexander Lomov
6. Hello, MongoDB Diaspora Expedience
8. The Problem ?
9. Consensus Problem
11. Consensus 1. Agreement 2. Integrity 3. Termination 4. Validity
12. Fail types node failure vs. network partition
13. Two Generals' Problem
14. Byzantine Generals Problem
15. Thus you need to have a minimum of 3m+1 processes to tolerate m faulty processes In other words more than 2/3’s need to be working – hence a minimum of 4 100% reliability requires infinite processes Lamport (1982) showed that in general agreement can be reached but:
16. Impossibility of Distributed Consensus with One Faulty MICHAEL J. FISCHER • Dijkstra award • Most influential papers in distributed computing • FLP result • Term “Single Point of Failure”
17. FLP Impossibility Fischer, Lynch and Paterson theorem proved that consensus in a fully asynchronous system cannot be guaranteed within bounded time
18. CAP Theorem
19. What solutions do we have ?
20. Quorum
21. Paxos
22. Raft Good Link with demo
23. Gossip
24. In Details
25. What else should be mentioned ?
26. Microservices FTW
27. Don’t overcomplicate
28. Performance vs. instance count
29. Fallacies of distributed computing The Fallacies of Distributed Computing are a set of assumptions that L. Peter Deutsch and others at Sun Microsystems originally asserted programmers new to distributed applications invariably make. [link]
30. The network is reliable. Latency is zero. Bandwidth is infinite. The network is secure. Topology doesn't change. There is one administrator. Transport cost is zero. The network is homogeneous. Fallacies of distributed computing
31. Call me Maybe https://aphyr.com/tags/Distributed-Systems
32. Home Reading 1. http://the-paper-trail.org/ 2. https://aphyr.com 3. http://kellabyte.com 4. http://www.bailis.org/ 5. http://www.neilconway.org/ 6. https://everythingisdata.wordpress.com/
34. Thank you Do you have any questions?