Garbage Collection in Java: Progress Since JDK 8

Since JDK 8, Java's garbage collection (GC) has undergone significant evolution, addressing common challenges like latency, pause times, and memory overhead. This article explores these advancements, focusing on practical implications for developers transitioning from older versions like JDK 8 to modern alternatives such as JDK 17 and JDK 21. Whether you’re maintaining legacy applications or planning future migrations, understanding these updates is crucial.

Key Points

1. Improvements Since JDK 8: Newer versions of the JDK offer significant enhancements in memory management and application performance.
2. Understanding GC Options: Choosing the right garbage collector for your application can optimize behavior and resource usage.
3. Incremental Updates: Advancements like generational GC modes and region-based heap layouts have transformed garbage collection, providing better scalability and efficiency.

Garbage Collection (GC) in Java automates memory management, freeing developers from handling low-level details. The two primary goals of GC are:

1. Fast Allocations: Java uses Thread-Local Allocation Buffers (TLABs) for fast, synchronization-free memory allocations.
2. Efficient Reclamation: GC algorithms reclaim unused memory through techniques like compaction and free lists.

Modern Java GC divides the heap into two generations:

• Young Generation: Stores short-lived objects, where collections are frequent but fast.
• Old Generation: Stores long-lived objects that survive multiple GC cycles.

This division is based on the generational hypothesis, which posits that most objects die young, making young generation collections more efficient than full heap collections. Java provides several GC algorithms, each tailored to specific use cases:

Garbage Collector

Focus

Use Case

Pause Time

Throughput

Garbage Collector	Focus	Use Case	Pause Time	Throughput
Serial GC	Low memory overhead	Small containers	Medium	Low
Parallel GC	High throughput	Batch processing or large datasets	High	High
G1 GC	Balanced performance	General-purpose, low-latency workloads	Medium-Low	Medium-High
ZGC	Ultra-low latency	Large-scale applications, low latency	Sub-millisecond	Medium
Shenandoah GC	Low latency	Large heaps, near-real-time processing	Very low	Medium

Serial GC

Low memory overhead

Small containers

Medium

Low

Parallel GC

High throughput

Batch processing or large datasets

High

High

G1 GC

Balanced performance

General-purpose, low-latency workloads

Medium-Low

Medium-High

ZGC

Ultra-low latency

Large-scale applications, low latency

Sub-millisecond

Medium

Shenandoah GC

Low latency

Large heaps, near-real-time processing

Very low

Medium

Introduced as the default collector in JDK 9, G1 GC uses a region-based heap layout and supports concurrent marking. This allows it to determine liveness without halting application threads. By combining young and old generation collections into smaller mixed collections, G1 reduces pause times and improves overall responsiveness.

Designed for ultra-low latency, ZGC can handle terabyte-sized heaps with pause times in the sub-millisecond range. It performs most of its work concurrently with application threads, making it ideal for applications requiring consistent responsiveness, such as cloud services or financial systems.

ZGC Generational Mode (introduced in JDK 21) further improves throughput by applying the generational hypothesis to separate short-lived and long-lived objects.

Benchmarks such as SPECjbb 2015 demonstrate substantial improvements in both throughput and latency across modern GC algorithms since JDK 8:

• Parallel GC: 30% improvement in throughput from JDK 8 to JDK 17.
• G1 GC: Over 40% improvement in throughput from JDK 8 to JDK 17.
• ZGC: 10% improvement with the generational mode in JDK 21.

Reduced Pause Times

Pause times have been drastically reduced across all collectors:

• Parallel GC: From ~100ms to ~65ms.
• G1 GC: 40% reduction from JDK 8 to JDK 17.
• ZGC: Sub-millisecond pauses.

G1 GC has seen significant reductions in native memory overhead, thanks to optimizations in remembered sets, data structures used for region-based collections. From JDK 8 to JDK 17, G1's native memory usage was cut almost in half. To better illustrate the practical aspects of GC, consider the following examples:

Example 1: Configuring G1 GC

# Add these options to your JVM startup command
java -XX:+UseG1GC -XX:MaxGCPauseMillis=50 -Xmx2g -Xms2g -jar app.jar

This configuration:

• Activates the G1 GC.
• Sets a target maximum pause time of 50ms.
• Allocates 2GB of heap memory.

Tuning ZGC for Low-Latency Applications

java -XX:+UseZGC -Xms4g -Xmx4g -XX:SoftRefLRUPolicyMSPerMB=50 -jar app.jar

This setup:

• Uses ZGC for ultra-low latency.
• Allocates 4GB of heap memory.
• Adjusts the lifetime of soft references for better memory management.

Challenges of Migrating Beyond JDK 8

While upgrading from JDK 8 to a newer version (e.g., JDK 17 or 21) can bring significant benefits, developers must consider:

• Compatibility Issues: Certain libraries or frameworks may not fully support newer JDK versions.
• Performance Tuning: Each GC has specific tuning parameters that may require adjustment for optimal performance.
• Staging Environment Testing: Always test thoroughly in non-production environments before rolling out changes.

The progress in Java's garbage collection since JDK 8 has been remarkable. With significant improvements in throughput, latency, and memory overhead, upgrading to newer JDK versions is necessary for any Java application.

Whether you're running small containers or large-scale cloud services, there's a GC algorithm optimized for your use case. So, if you're still on JDK 8, it's time to make leap and enjoy the performance benefits of modern Java.

For more information, watch this video from Devoxx Belgium about Garbage Collection in Java: The Progress Since JDK 8 by Stefan Johansson

?

The above is the detailed content of Garbage Collection in Java: Progress Since JDK 8. For more information, please follow other related articles on the PHP Chinese website!