bottleneck
The single most constrained part of a system or process, which limits how fast the whole thing can go.
In more detail
Speeding up anything else makes little difference, because work just queues at the bottleneck. Measure first to find it, fix it, then measure again, because the bottleneck moves.
Related terms
- profiling — Measuring where a program spends its time and memory while it runs a representative operation.
- saturation — How full a system’s most constrained resource is, such as its CPU, memory, connections or queue.
- N+1 query problem — A performance bug where code runs one query for a list and then one more query per item, so N items cost N+1 queries.
- hot partition — A partition that gets far more traffic than the others because many requests use the same key, causing throttling even when total capacity is free.
- value stream map — A diagram or table of every step one change goes through from request to production, showing both working time and waiting time.
- wait time — Time a piece of work spends sitting in a queue, waiting for someone or something, with nobody working on it.
- leading and lagging indicators — Lagging indicators report what has already happened; leading indicators move early and predict what will happen.
Taught in
- Load Testing — Modern API Architecture: REST, GraphQL & gRPC
- Performance Optimization — Modern API Architecture: REST, GraphQL & gRPC
- Database Performance: Caches, Proxies And Capacity Units — AWS Certified Solutions Architect – Associate (SAA-C03) Exam Prep
- The Performance Clinic: Five Slow Systems, Five Fixes — AWS Certified Solutions Architect – Associate (SAA-C03) Exam Prep
- The Pitfalls Of Measurement — DevOps & Site Reliability Engineering
- Partitioning, Sharding And Hot Keys — Scalable System Design & Distributed Architecture
References
- Google SRE book: Monitoring distributed systems
- Brendan Gregg: The USE Method
- DORA's software delivery performance metrics