Mastering JVM Performance Tuning: Practical Steps and Real‑World Examples

After covering the basics of the JVM series, the goal shifts to preparing for JVM performance tuning. Performance tuning spans multiple layers—architecture, code, JVM, database, and OS—where architecture and code tuning form the foundation, with JVM tuning as a last resort.

When to Perform JVM Tuning

Heap memory (old generation) continuously grows to its maximum limit.

Frequent Full GC occurrences.

GC pause times exceed one second.

Application throws OutOfMemoryError or other memory exceptions.

Heavy use of native caches consuming large memory.

System throughput or response performance degrades.

Basic Principles of JVM Tuning

Most Java applications do not need JVM tuning.

Most GC issues stem from code-level problems.

Set optimal JVM parameters before deployment.

Reduce object creation at the code level.

Avoid excessive global variables and large objects.

Prioritize architecture and code tuning; JVM tuning is a fallback.

Analyzing GC logs and optimizing code is more effective than tweaking JVM flags.

The most effective optimizations are therefore at the architecture and code levels, with JVM tuning serving as the final "squeeze" of server configuration.

JVM Tuning Objectives

The ultimate aim is to achieve the smallest hardware footprint while delivering higher throughput. JVM tuning focuses on garbage collector performance to reduce memory usage and latency while increasing throughput.

Latency: low GC pause time and low GC frequency.

Low memory consumption.

High throughput.

Improving one metric often compromises another; targets should be set based on business priorities.

Quantitative Tuning Targets

Heap usage ≤ 70%.

Old generation usage ≤ 70%.

Average GC pause ≤ 1 s.

Full GC count = 0 or average pause interval ≥ 24 h.

These targets vary per application.

JVM Tuning Steps

Analyze GC logs and heap dumps to identify bottlenecks.

Define quantitative tuning goals.

Determine JVM parameters based on historical settings.

Iteratively tune memory, latency, and throughput metrics.

Compare pre‑ and post‑tuning performance.

Repeat analysis and adjustment until an optimal configuration is found.

Apply the final parameters to all servers and monitor continuously.

Some steps require multiple iterations, typically starting with memory requirements, then latency, and finally throughput.

Key JVM Parameters

JVM tuning relies heavily on command‑line flags. -XX options are considered "unstable" and can cause significant performance variations; proper usage can greatly improve stability.

Boolean options: -XX:+Option (enable) or -XX:-Option (disable).

Numeric options: -XX:Option=Number (supports units like k, m, g).

String options: -XX:Option=String (e.g., -XX:HeapDumpPath=./dump.core).

Parameter Examples and Interpretation

-Xmx4g -Xms4g -Xmn1200m -Xss512k -XX:NewRatio=4 -XX:SurvivorRatio=8 -XX:PermSize=100m -XX:MaxPermSize=256m -XX:MaxTenuringThreshold=15

Explanation: -Xmx4g: maximum heap size 4 GB. -Xms4g: initial heap size 4 GB. -Xmn1200m: young generation size 1200 MB (≈3/8 of total heap). -Xss512k: thread stack size 512 KB (adjust based on thread count). -XX:NewRatio=4: young vs. old generation ratio 1:4. -XX:SurvivorRatio=8: Eden to Survivor ratio 8:1. -XX:PermSize/MaxPermSize: permanent generation sizes (obsolete after Java 8). -XX:MaxTenuringThreshold=15: max age before promotion to old generation.

Other useful flags include -XX:+DisableExplicitGC, -XX:CMSInitiatingOccupancyFraction=60, -XX:ConcGCThreads, -XX:ParallelGCThreads, and various CMS/Full GC controls.

Memory Optimization Example

After a stable run, a Full GC log shows old generation usage of ~93 MB. Based on this, the following scaling rules are applied:

Heap ( -Xms / -Xmx) ≈ 3–4 × old‑gen usage.

Permanent generation ( -XX:PermSize / -XX:MaxPermSize) ≈ 1.2–1.5 × old‑gen usage.

Young generation ( -Xmn) ≈ 1–1.5 × old‑gen usage.

Old generation ≈ 2–3 × old‑gen usage.

Resulting command:

java -Xms373m -Xmx373m -Xmn140m -XX:PermSize=5m -XX:MaxPermSize=5m

Latency Optimization Example

Key latency metrics include average and maximum pause times, Minor GC frequency, and Full GC frequency. Reducing young generation size shortens Minor GC pause but increases its frequency, and vice‑versa. By decreasing young generation by 10 %:

java -Xms359m -Xmx359m -Xmn126m -XX:PermSize=5m -XX:MaxPermSize=5m

Throughput Optimization

Throughput tuning aims to minimize Full GC and Stop‑The‑World events. Strategies include keeping most reclamation in Minor GC, preventing rapid promotion to old generation, and adjusting heap sizes to maintain a balance between memory usage and GC overhead.

Tools

GCViewer can visualize GC logs, helping to assess total heap, tenured heap, young heap usage, GC pause times, and frequencies. Lower values generally indicate better performance.

References:

https://blog.csdn.net/jisuanjiguoba/article/details/80176223

https://juejin.im/post/59f02f406fb9a0451869f01c