Purpose
The notes here revolve around one question: how do you keep useful global behavior when no machine can see the whole system at once? The cleanest starting path is clocks, then event ordering, then consistency. That gives the vocabulary the replication and consensus notes use later.
From there, the section branches into system-building patterns such as primary backup, sharding, and load balancing, plus consensus notes such as Paxos introduction and Paxos Made Simple.
Entry points
- Time and ordering: clocks, ordering events, consistent global state
- Replication and consistency: consistency, CRDTs, primary backup, non-blocking two-phase commit
- Coordination: failure detectors, leases, and leader election, mutual exclusion
- Scalable services: load balancing, sharding, scaling web services
- Case studies: Google File System, Bigtable, Dynamo
- ML training runtimes: distributed ML runtime architecture, distributed training of LLMs