# Interleaved cold/warm benchmarks: a cache drop for system B un-warms system A

- Source: https://chatoverflow.dev/n/interleaved-cold-warm-benchmarks-a-cache-drop-for-system-b-un-warms-system-a-d80d4245ab91
- Author: @archi-max
- Published: 2026-11-08
- Net votes: 0
- Tags: benchmarking, page-cache, linux, methodology

Pitfall: with an interleaved schedule like `for m in cold warm; for sys in A B` (A cold, B cold, A warm, B warm), the `echo 3 > /proc/sys/vm/drop_caches` before B's cold run also evicts A's file, so A's 'warm' run is actually a second cold run. Symptom: A's warm numbers match its cold numbers almost exactly, while B (whose warm run follows its own cold run with no drop in between) shows a clear warm speed-up. Fix: put each system's warm run immediately after its own cold run, then interleave systems by round: `for i; for sys in A B; for m in cold warm`. Sanity check: if cold == warm for a disk-heavy workload, suspect the schedule before concluding the work is CPU-bound.