<![CDATA[This study evaluates the performance characteristics of a real-time WebSocket and Redis architecture deployed on a resource-constrained Virtual Private Server (1 vCPU, 4 GB RAM) using a GitHub Actions CI/CD pipeline and Docker Compose. Three configurations are tested: C1 as a baseline without Redis pub/sub, C2 with Redis pub/sub enabled, and C3 combining Redis pub/sub with a fully warmed HTTP cache. Load tests using Grafana K6 simulate 500 concurrent virtual users. Results show the server sustains 335 simultaneous WebSocket connections with zero real errors. Redis pub/sub increases average broadcast delivery latency from 23.78 ms to 42.05 ms and introduces a 1.65% spike rate, caused by O(N) broadcast fan-out on the Node.js libuv event loop rather than Redis itself, whose round-trip latency remained sub-millisecond. Redis HTTP caching reduces cache-miss response time by 39.6–51.4%. JMeter’s eu.luminis plugin proved incompatible with wss:// endpoints, returning HTTP 400 for all 902 connections, motivating adoption of Grafana K6. This work contributes quantitative measurement of Redis pub/sub overhead trade-offs in single-core constrained WebSocket architectures with automated CI/CD performance gates]]>
Copyrights © 2026
