RWS Scalability Benchmark Report

RWS memory stability follow-up

A separate cold-start test measured 11.11 MiB RSS after 20 million static keep-alive requests. This is a stability measurement under different conditions, not a direct memory comparison with Nginx or Apache. Read the full memory stability report.

Scope and validity

Requested and achieved connection counts are kept separate. Server metrics are collected independently from load-generator measurements. A warning means the corresponding result must not be treated as an unconstrained capacity claim.

Environment and method

The offered rate capped RWS and Nginx near 10,000 RPS, so this experiment compares behavior at that load and does not establish either server's maximum throughput.

Individual runs

Server resources from the original comparison

These are historical per-run measurements from the comparison above; the separate 20-million-request follow-up is not included in this table.

Comparison

Aggregation: median repeated runs.

Relative to apache

• nginx: throughput +264.09%; p99 latency -13.95%.
• rws: throughput +265.31%; p99 latency -25.38%.

Interpretation

• RWS and Nginx both sustained more than 99% of the offered rate in all accepted runs with zero failed responses.
• RWS median p99 latency was 7.10 ms, 13.3% lower than Nginx at 8.19 ms. Median throughput differed by only 0.33% because both were constrained by the 10,000 RPS offer.
• Apache saturated well below the offered rate: median throughput was 2,723.67 RPS and the three runs recorded 250 failed responses in total. Latency percentiles include only completed responses and therefore understate Apache's failure impact.
• In a separate, later cold-start stability test, RWS measured 11.11 MiB RSS after 20 million static keep-alive requests. This result is not directly comparable to the historical Nginx and Apache resource measurements above.
• The 100 ms CPU sampler has 10 percentage-point resolution on this one-vCPU host; CPU figures are directional rather than precise enough for small percentage claims.

Warnings and deterministic analysis

apache: Achieved response rate was below 90% of the configured offered rate; inspect load-generator saturation and errors before attributing the result to the server.

apache: achieved throughput is below 90% of offered load; the result is saturation or generator-limited, not an unconstrained throughput figure.

apache: socket errors occurred; successful-request latency alone is not representative.

apache: Achieved response rate was below 90% of the configured offered rate; inspect load-generator saturation and errors before attributing the result to the server.

apache: achieved throughput is below 90% of offered load; the result is saturation or generator-limited, not an unconstrained throughput figure.

apache: socket errors occurred; successful-request latency alone is not representative.

apache: Achieved response rate was below 90% of the configured offered rate; inspect load-generator saturation and errors before attributing the result to the server.

apache: achieved throughput is below 90% of offered load; the result is saturation or generator-limited, not an unconstrained throughput figure.

apache: socket errors occurred; successful-request latency alone is not representative.

Charts

Raw data

The source JSON retains settings, categorized errors, resource samples and mergeable HDR-style latency buckets. No AI or external API is used for calculations.