Summary
Lowest-p99 winner per workload & concurrency, with headline stats. All comparisons are relative on identical hardware.
Winner heatmap — lowest p99 latency
Each cell shows the server with the lowest median p99 at that concurrency.
Memory
Peak resident set size (RSS) per server — the max observed across all workloads.
Peak RSS per server
Lower is better. Persistent workers (Octane) carry the framework in memory; FPM forks per request.