GC Algorithms at a Glance

A quick field guide to the collectors EasyGC recognizes, and when to pick which.

The collectors

Collector	Flag	Strategy	Typical pause	Good for
Serial	-XX:+UseSerialGC	Single-threaded, generational	100 ms – seconds	Small heaps, single-core containers, batch tools
Parallel	-XX:+UseParallelGC	Multi-threaded, generational	50 ms – seconds	Max throughput, batch/analytics, latency-tolerant
CMS	-XX:+UseConcMarkSweepGC	Mostly concurrent old-gen	10 – 100 ms	Legacy low-latency services (removed in Java 14)
G1	-XX:+UseG1GC	Region-based, concurrent marking	10 – 200 ms (targeted)	Default since Java 9; balanced throughput/latency
ZGC	-XX:+UseZGC	Concurrent, colored pointers	< 1 ms	Huge heaps, strict SLOs (Java 15+)
Shenandoah	-XX:+UseShenandoahGC	Concurrent, forwarding pointers	< 10 ms	Low latency on OpenJDK builds

How to choose

1. Default choice: G1. It is the JDK default for a reason — predictable pauses with decent throughput.
2. Throughput above all (batch jobs, data pipelines)? Parallel GC gives the best raw throughput at the cost of longer pauses.
3. Strict latency SLO (p99 < 50 ms) with a large heap? ZGC or Shenandoah keep pauses in the sub-10 ms range regardless of heap size.
4. Tiny container (≤ 512 MB)? Serial GC often beats everything — no threads to coordinate.

Key tuning knobs

Knob	Meaning
-Xms / -Xmx	Initial / maximum heap size. Setting them equal avoids resize pauses.
-Xmn / -XX:NewRatio	Young generation size — the main lever on GC frequency.
-XX:MaxGCPauseMillis	G1 pause goal (default 200 ms). Don't set it unrealistically low.
-XX:MaxMetaspaceSize	Cap for class metadata — prevents metaspace-driven Full GCs.
-XX:+DisableExplicitGC	Ignore System.gc() calls (check Direct ByteBuffers first!).
-XX:G1HeapRegionSize	G1 region size — raise it if humongous allocations dominate.

Reading the signals in your report

• Allocation Failure young GCs are normal and healthy.
• System.gc() causes are almost always worth removing.
• Metadata GC Threshold / Metaspace causes point at classloader churn.
• G1 Humongous Allocation means objects larger than half a region — refactor or enlarge regions.
• A steadily climbing after-GC heap means leaked objects survived collection — see Understanding the Report.