L Peter Deutsch: The Fallacies of Distributed Computing

Snippet from: The Wikipedia Article, along with my own notes

  • The network is reliable : Think about error-handling on network failures. Use the following patterns: Timeouts, Retries, Circuit breakers, Graceful Degradation.
  • Latency is zero : Think about bandwidth requirements and packet loss. Use the following patterns: Small payloads, Compression, Asynchronous Communication, Perceived Progress, Timeouts, Graceful Degradation.
  • Bandwidth is infinite : The same as above. Think explicitly about bottlenecks in Network traffic. Use the following patterns: Caching, Etags.
  • The network is secure : Think about encryption, compliance, data risk. Understand the trade-offs with speed and simplicity. Prioritize security over the network for any sensitive data.
  • Topology doesn't change : Think about local state, tolerance for staleness, co-ordination overheads, recovery mechanisms for split brains, inability to reach endpoints.
  • There is one administrator : Remember to test the end-to-end flow and think through production components not controlled by you.
  • Transport cost is zero : Think through data transfer costs in cloud environments. Monitor payload sizes.
  • The network is homogeneous : Think about connection drops, reconnects, client timeouts. Use the following patterns: Idempotency, Eventual Consistency

Last Updated: 2024-04-17


Tags