The Computer Science behind a modern distributed data store
summary
Explains concrete systems and tradeoffs: consensus uses Raft requires an odd-numbered quorum and majority replication to append a persisted log and elect a leader; modern sorting salvages merge heuristics to exploit 20,000× compute vs 40× memory and 32 cores with a min-heap that achieves 95% cache hits for merges; log-structured merge trees plus bloom filters yield sequential writes and fast hot-set reads; hybrid logical clocks take max(local, seen) timestamps; distributed ACID needs multi-versioning plus replicated commit metadata.
AI-written summary · about
chapters (13)
- 0:00Intro
- 0:26Topics
- 1:24Resilience Consensus
- 3:53Paxos Consensus
- 6:26Why its odd
- 9:54Raft
- 11:23Sorting
- 14:00Merging
- 18:53Lock structured merge trees
- 21:03Log structured merge stream
- 27:23Summary
- 28:30Hybrid Logical Clock
- 47:56Questions
related talks
47mDistributed Systems: When Democracy and Computer Science Come Together
45mDo Distributed Systems they said. It'll be fun they said.
50mDo Distributed Systems they said. It'll be fun
40mConsensus Algorithms in Distributed Systems
14mDistributed Consensus in 15 Minutes!
44mCluster Consensus: When Aeron Met Raft
13mThe ACID BASE Conundrum
22mWhy Are Distributed Systems so Hard?