Jvm Performance

JVM Performance Optimization

Java Virtual Machine performance tuning, garbage collection, and profiling techniques. Use when diagnosing performance issues, tuning JVM parameters, or optimizing Java applications.

Note: This document synthesizes JVM performance concepts from various sources including Oracle documentation, JVM specifications, and community best practices.

Performance Trade-offs

JVM tuning involves inherent trade-offs. Improving one metric often impacts another.

2. Garbage Collection Algorithms

See references/gc-tuning.md for detailed GC algorithms (mark-and-sweep, generational collection, STW pauses), JVM tuning parameters, memory analysis, and JIT compilation.

3. Garbage Collectors

See references/garbage-collectors.md for detailed information on:

• Serial GC, Parallel GC, G1 GC
• ZGC (Z Garbage Collector)
• Shenandoah

4. GC Selection Guide

5. Performance Analysis Approach

Systematic Process

1. Define performance goals with specific metrics
2. Measure baseline performance
3. Identify bottlenecks through profiling
4. Make targeted changes
5. Verify improvement with measurements
6. Document findings

Measurement Principles

• Statistical significance: Multiple runs required
• Control environment: Same hardware, data, load
• Measure before and after: Quantify change impact
• Non-normal distributions: Use percentiles, not just means

7. Profiling Tools

JDK Flight Recorder (JFR)

Low-overhead production profiling.

# Start recording
jcmd <pid> JFR.start name=profile duration=60s filename=recording.jfr

# Or via JVM args
-XX:StartFlightRecording=duration=60s,filename=recording.jfr

Events:

• CPU usage
• Memory allocation
• GC events
• Thread events
• Method profiling

Java Mission Control (JMC)

GUI for JFR analysis.

Key Views:

• Event browser
• Thread analysis
• Memory analysis
• Code profiling

async-profiler

Low-overhead sampling profiler.

# CPU profiling
./profiler.sh -d 60 -f cpu.html <pid>

# Allocation profiling
./profiler.sh -d 60 -e alloc -f alloc.html <pid>

JMX Monitoring

# Enable remote JMX
-Dcom.sun.management.jmxremote
-Dcom.sun.management.jmxremote.port=9010
-Dcom.sun.management.jmxremote.authenticate=false
-Dcom.sun.management.jmxremote.ssl=false

8. Common Performance Mistakes

Optimizing Without Measurement

Making changes based on assumptions rather than data.

Warning signs:

• Complex code for assumed performance
• No profiling data
• "It feels faster"

Copy-Paste Tuning

Applying JVM flags without understanding their impact.

Warning signs:

• Copy-paste JVM flags from blogs
• Using outdated tuning advice
• Ignoring workload characteristics

Flawed Microbenchmarks

Microbenchmarks can produce misleading results.

Common issues:

• JIT compilation effects
• Dead code elimination
• Warmup not considered

Use JMH for microbenchmarks:

@BenchmarkMode(Mode.AverageTime)
@OutputTimeUnit(TimeUnit.NANOSECONDS)
@Warmup(iterations = 3, time = 1)
@Measurement(iterations = 5, time = 1)
@Fork(1)
public class MyBenchmark {
    
    @Benchmark
    public void testMethod() {
        // benchmark code
    }
}

Ignoring Tail Latency

Average latency can hide problematic outliers.

Wrong: Average is 50ms Right: P99 is 500ms, indicates tail latency problem

9. Virtual Threads (Java 21+)

Benefits

• Lightweight (millions possible)
• No thread pool management
• Simpler async code

When to Use

Good fit:

• I/O-bound workloads
• Many concurrent tasks
• Blocking APIs

Not good fit:

• CPU-bound tasks
• Synchronized blocks (pins carrier thread)
• Thread-local heavy code

Implementation

// Create virtual thread
Thread.startVirtualThread(() -> {
    // Task code
});

// ExecutorService with virtual threads
try (var executor = Executors.newVirtualThreadPerTaskExecutor()) {
    IntStream.range(0, 10000).forEach(i -> {
        executor.submit(() -> {
            Thread.sleep(Duration.ofSeconds(1));
            return i;
        });
    });
}

Pinning Issues

Pinning causes: Carrier thread blocked, reducing throughput.

Avoid:

• synchronized blocks/methods
• Native methods that block

Fix:

// Replace synchronized with ReentrantLock
// Before
synchronized(lock) { ... }

// After
private final ReentrantLock lock = new ReentrantLock();
lock.lock();
try { ... } finally { lock.unlock(); }

10. Cloud-Native Considerations

Container Memory Limits

# JVM respects container limits (Java 10+)
-XX:+UseContainerSupport

# Limit JVM heap to leave room for off-heap
# Rule: Heap = Container Memory * 0.75 - Off-heap estimate
-Xmx6g  # In 8GB container with ~1GB off-heap

Startup Optimization

# Class Data Sharing
java -Xshare:dump
-XX:+UseSharedSpaces

# AOT compilation (GraalVM)
native-image -jar app.jar

# CDS with dynamic archive
-XX:ArchiveClassesAtExit=app.jsa
-XX:SharedArchiveFile=app.jsa

Observability Stack

┌─────────────┐
│ Application │
│   (JVM)     │
└──────┬──────┘
       │
       ▼
┌─────────────┐     ┌─────────────┐     ┌─────────────┐
│ Micrometer  │────▶│ Prometheus  │────▶│  Grafana    │
│  (metrics)  │     │  (storage)  │     │ (dashboard) │
└─────────────┘     └─────────────┘     └─────────────┘
       │
       ▼
┌─────────────┐     ┌─────────────┐
│OpenTelemetry│────▶│    Jaeger   │
│  (tracing)  │     │   (traces)  │
└─────────────┘     └─────────────┘

11. Performance Troubleshooting Checklist

High CPU Usage

1. Profile with async-profiler
2. Check for GC overhead (jstat -gcutil)
3. Look for busy loops
4. Check for excessive logging

High Memory Usage

1. Check heap usage (jcmd GC.heap_info)
2. Look for memory leaks (heap dump)
3. Analyze GC logs
4. Check Metaspace for class leaks

Long GC Pauses

1. Check GC logs (-Xlog:gc*)
2. Analyze pause times vs goals
3. Consider different GC algorithm
4. Check for heap sizing issues

Slow Startup

1. Profile with JFR
2. Check class loading (-Xlog:class+load)
3. Consider CDS or AOT
4. Reduce classpath scanning

JVM Flags Quick Reference

# Essential logging
-Xlog:gc*:file=gc.log:time,uptime,level,tags

# Memory
-Xms4g -Xmx4g
-XX:MetaspaceSize=256m -XX:MaxMetaspaceSize=512m

# G1 GC
-XX:+UseG1GC
-XX:MaxGCPauseMillis=200

# ZGC (Java 15+)
-XX:+UseZGC
-XX:ZCollectionInterval=0

# Flight Recorder
-XX:StartFlightRecording=duration=60s,filename=rec.jfr

# Heap dump on OOM
-XX:+HeapDumpOnOutOfMemoryError
-XX:HeapDumpPath=/tmp/heap.hprof

Related Skills

• spring-framework: Actuator, Micrometer setup, Spring-specific debugging
• k8s-workflow: Container resource management
• dockerfile: JVM containerization patterns

References

• Oracle JVM Documentation
• Java Performance by Scott Oaks (O'Reilly)
• GC Handbook by Charlie Hunt, Binu John
• Optimizing Java (2nd Edition) by Benjamin Evans, James Gough (for deeper study)