Distribution changes the rules

What changes when work crosses a network?

1.1From one server to a distributed serviceNodes, clients, local state, and messages in one interaction.

1.2

The network is uncertain

Latency, timeouts, retries, and ambiguous failure.

Planned

Copies disagree

What happens when the same data lives in several places?

2.1Replicate the dataAdd a second copy and see what proximity changes.2.2Follow a writeCommit first, acknowledge, then replicate.2.3Read from a replicaRoute a client to the nearest available copy.2.4Observe a stale readRace a read against replication lag.

2.5

Choose a consistency promise

Compare eventual, session, and linearizable reads.

Planned

2.6

Wait for a quorum

Trade latency and availability for stronger agreement.

Planned

Agree through failure

How do independent nodes commit one history?

3.1

Order concurrent writes

See why clocks cannot settle every conflict.

Planned

3.2

Commit one history

Replicated logs, majorities, and consensus.

Planned

3.3

Split the network

Partitions, split brain, and the CAP tradeoff.

Planned

3.4Recover from failureDetect, elect, repair, and resume one action at a time.

Distribute the workload

How does the design change when one copy is not enough?

4.1

Partition the data

Divide keys between nodes and route each request.

Planned

4.2

Rebalance the cluster

Move ownership safely when nodes join or leave.

Planned

4.3

Handle a hotspot

Respond when one key receives most of the traffic.

Planned

4.4

Coordinate across shards

Atomicity, two-phase commit, and partial completion.

Planned

See the system make every decision.

Jump to a lesson

Distribution changes the rules

Copies disagree

Agree through failure

Distribute the workload