NOTE

Benchmarking Elasticsearch

Representative search/index benchmarks, warm-up, dataset/query realism, latency percentiles, and bottleneck measurement.

Elasticsearch / SearchCreated 05/15/2022Updated 09/15/20261 min readhistorical

This is a historical learning note and may contain outdated or incomplete understanding.

Benchmark with a representative corpus, mapping, query distribution, ingestion rate, concurrency, shard layout, and hardware/storage environment.

Measure throughput together with p50/p95/p99 latency, indexing/search errors, rejections, GC, CPU, disk I/O, cache hit behavior, merge pressure, and queueing.

Warm-up and steady-state matter because caches/JIT/segment state change early results. Tools such as Rally-style benchmarks are useful, but production-shaped custom tracks are more valuable than generic headline QPS.