Common JVM Garbage Collectors: Characteristics, Use Cases, and Tuning Parameters

Java garbage collection is a frequent interview topic and an essential knowledge area for developers. This article introduces the main JVM garbage collectors, their key features, appropriate scenarios, and tuning suggestions.

Young Generation Collectors

Serial

Serial is the oldest collector, using a single thread and a copying algorithm. It pauses all application threads during collection.

Target: Young generation

Algorithm: Copying

Threading: Single‑threaded

Pause: Stops all user threads

Advantages

Simple and efficient on single‑CPU systems because there is no thread‑switching overhead.

Disadvantages

Stops all application threads, which may be noticeable to users.

Usage Scenarios

Client‑side desktop applications with limited memory

Single‑core servers where maximum single‑thread throughput is desired

Parameter

-XX:+UseSerialGC

: Explicitly enable the Serial collector

ParNew

ParNew is the multithreaded version of Serial, sharing the same algorithm but using multiple threads.

Advantages

Better utilization of multi‑core CPUs.

Disadvantages

Still pauses all application threads during collection.

Usage Scenarios

Used in Server mode; it is the only young‑generation collector that can work together with the CMS collector.

Parameters

-XX:+UseConcMarkSweepGC

: When CMS is enabled, ParNew becomes the default young‑generation collector -XX:+UseParNewGC: Force the use of ParNew -XX:ParallelGCThreads: Set the number of GC threads (default equals the number of CPUs)

Parallel (Parallel Scavenge)

Parallel Scavenge is a multithreaded young‑generation collector that also uses the copying algorithm. Its goal is to achieve a controllable throughput rather than minimizing pause time.

Target: Young generation

Algorithm: Copying

Threading: Multithreaded

Focus: High throughput, predictable pause time

Advantages

High throughput, efficient CPU usage, suitable for workloads that can tolerate longer pauses.

Disadvantages

Higher throughput comes at the cost of longer pause times compared with pause‑time‑focused collectors.

Usage Scenarios

Ideal for applications running on multi‑CPU servers where pause time is not critical, such as batch processing, order handling, payroll, or scientific calculations.

Parameters

-XX:MaxGCPauseMillis

: Target maximum pause time (ms) -XX:GCTimeRatio: Desired GC time as a percentage of total runtime (0‑99)

Old Generation Collectors

Serial Old

Serial Old is the old‑generation counterpart of Serial, using a single thread and the mark‑compact algorithm.

Target: Old generation

Algorithm: Mark‑compact (also called Mark‑Sweep‑Compact)

Threading: Single‑threaded

Usage Scenarios

Client mode

Single‑core servers

Paired with Parallel Scavenge

Fallback for CMS when concurrent mode fails

CMS (Concurrent Mark‑Sweep)

CMS aims to minimize pause times by performing most of its work concurrently. It uses the mark‑sweep algorithm.

Phases: Initial mark, concurrent marking, remark, concurrent sweep

Target: Old generation

Algorithm: Mark‑sweep (no compaction, may cause fragmentation)

Goal: Short pause times, high throughput

Drawback: Higher memory consumption and CPU sensitivity

Advantages

Short pause times

High throughput

Concurrent collection

Disadvantages

CPU‑intensive

Cannot collect floating garbage

May cause significant memory fragmentation

Usage Scenarios

Interactive applications that require low latency

Web/B/S server‑side services where response time is critical

Parameter

-XX:+UseConcMarkSweepGC

: Enable CMS collector

Parallel Old

Parallel Old is the old‑generation version of Parallel Scavenge, using multiple threads and the mark‑compact algorithm.

Target: Old generation

Algorithm: Mark‑compact

Threading: Multithreaded

Usage Scenarios

JDK 1.6+ as a replacement for Serial Old

Server mode on multi‑CPU machines where throughput is prioritized

Parameter

-XX:+UseParallelOldGC

: Enable Parallel Old collector

G1 (Garbage‑First)

Characteristics

Parallel and concurrent, fully utilizing multi‑CPU environments

Generational collection covering both young and old generations

Combines multiple algorithms to avoid fragmentation

Predictable pause times with configurable targets

Designed for server‑side workloads, intended to replace CMS

Advantages

Effective use of multi‑core hardware

Manages the entire heap without needing other collectors

No fragmentation, suitable for long‑running services

Provides a model for predictable pause times

Disadvantages

G1 requires a large remembered set (up to ~20% of heap) and incurs additional overhead, making it less efficient for small‑memory applications compared with CMS.

Usage Scenarios

Recommended for most modern server applications; for non‑performance‑critical cases, G1 can be used as the default collector.

Parameters

-XX:+UseG1GC

: Enable G1 collector -XX:InitiatingHeapOccupancyPercent: Start concurrent marking when heap usage reaches this percent (default 45) -XX:MaxGCPauseMillis: Target maximum pause time (default 200 ms) -XX:G1HeapRegionSize: Size of each region (1 MB‑32 MB, aiming for ~2048 regions at minimum heap)

Understanding these collectors and their tuning options equips developers to handle interview questions and to optimise Java applications for years to come.