JVM tuning for production servers

Deploying a Java application on a server with default JVM settings is like buying a sports car and only ever driving in second gear. Proper configuration can be the difference between a 100ms and a 5-second response time.

Heap size¶

Basic rule: set -Xms and -Xmx to the same value. Dynamic heap size changes cause full GC pauses. On servers with 16 GB of RAM we typically allocate 8–10 GB for the JVM heap. PermGen: 256–512 MB for enterprise applications.

Garbage Collection — CMS¶

For web applications we use the Concurrent Mark Sweep collector. Parallel GC has better throughput, but its stop-the-world pauses can last seconds. CMSInitiatingOccupancyFraction=70 — CMS starts at 70% fill of the old generation.

GC logging¶

GC logging is enabled on every production server. We analyze it with GCViewer. Target metrics: minor GC under 50ms, full GC less than once per hour, total GC time under 5%.

Thread stack and monitoring¶

Reducing -Xss to 512 KB saves memory for more threads. JMX monitoring via Nagios: heap utilization, thread count, GC activity. Alert when heap exceeds 85%.

Summary¶

Xms = Xmx. CMS for web applications. GC logging always enabled. Monitor via JMX. Test under load on staging before production.