Tune Code Before Your Garbage Collector
Summary
The post argues that optimizing memory allocations in Java can yield larger benefits than GC selection alone, and that logging overhead significantly impacts tail latency. It uses a 50K/s MarketDataSnapshot to NewOrderSingle benchmark (via JLBH and Chronicle-FIX) to compare SLF4J logging versus Chronicle-Queue logging, showing substantial p99 and p99.99 improvements when logging is minimized or moved, and highlights the role of disk IO placement. It concludes that workload tuning should come before GC choices, with practical guidance on building low-latency Java paths.