GC Algorithm

Parallel GC

Applied as the default GC in Java 8

• Performs minor collection in parallel

  • Can significantly reduce GC overhead

• A GC designed for applications that handle medium to large-scale datasets running on multi-processor or multi-thread hardware

• Parallel Compaction

  • A feature that allows the Parallel Collector to perform major collection in parallel

  • When the -XX:+UseParallelGC option is specified, Parallel Compaction is used by default

• However, the use of multi-threaded collection is limited to the Young generation only

  • Young Generation Collection algorithm: Parallel Scavenge

  • Old Generation Collection algorithm: Serial Mark-Sweep-Compact

G1 (Garbage First Collector)

Applied as the default GC in Java 11

• A Collector designed for multi-processor machines with large memory

• The G1 Collector can be enabled with the -XX:+UseG1GC option

• Note that G1 GC is the only one that is not a Generational GC

  • The G1 collector is a unique collector that does not divide the heap into Young and Old regions

  • G1 eliminates the physical generation distinction and treats the entire heap as 1MB-sized regions

    • It is organized in a grid pattern and uses approximately 2000 regions

  • The name G1 means that collection starts from the region most filled with garbage

    • When a region full of garbage is found, the collector runs immediately

• Live objects in Old regions are moved to other Old regions, and compaction occurs

• In G1, the concepts of Young and Old generations are not fixed

  • The set of regions where objects are newly allocated is the Young generation

  • The set of regions where promotion occurs is the Old generation

ZGC

Applied as the default GC in Java 17

ZGC is a scalable low-latency GC that can handle heaps ranging from 8MB to 16TB in size, and can be enabled using the -XX:+UseZGC option

• Low-latency

  • ZGC increases responsiveness by minimizing pauses for memory collection

  • It is a concurrent GC where application threads continue to run during collection operations, and pause times are very short

    • This allows maintaining high responsiveness even in large-scale memory environments

    • Maximum latency is less than 1 millisecond

• Very Large Heap Support

  • ZGC can handle heaps as large as 16TB

    • This makes it suitable for applications that perform large-scale data processing or memory-intensive tasks

• Non-contiguous Collection

  • Instead of collecting the entire heap at once, ZGC divides it into small region units and performs garbage collection

  • By processing only portions of the collection work, it can distribute pause times

• Parallel Processing

  • ZGC utilizes multi-threads to process GC operations in parallel

  • This efficiently distributes GC operations and reduces overall collection time

• Efficiency

  • It is efficient because there is no need to clean up the heap during collection

    • Typical GCs trigger stop-the-world events to perform memory collection operations

      • During this time, all application threads are paused for GC operations

      • These pause times can affect application responsiveness and performance

    • In contrast, ZGC maintains very short pause times during the heap cleanup process

  • ZGC continues to run application threads during collection operations and only triggers very short pauses when necessary

    • ZGC focuses on increasing application responsiveness by minimizing pause times