Why Garbage-First (G1) targeted for multiprocessor machines with large memories

Question

According to:

9 Garbage-First Garbage Collector

and:

G1: Java's Garbage First Garbage Collector

G1 targeted for multiprocessor machines with large memories.

Those 2 papers (and other web papers), does not describe why we need: a. large memories b. multiprocessor (I assume this need due to concurrent & parallel)

What is the technical explanation for those requirements ?

Answer 1

It's other way around. G1 is not targeted for large memories. If your application demands large heap size, G1 is effective.

Why your application demands large heaps? It's depend on business requirements and specific needs of application. You may load huge set of master data Or you may use in - memory caching to reduce response times. Think of big data applications,(Spark,Hadoop) which are processing teta bytes of data and use memory for processing.

Multiprocessors machines have more processing powers and effective for parallel execution of different tasks. Large heap applications obviously demands more processing power.

By setting Max pause time goal, G1GC try to meet that goal. Compared to other algorithms, by default G1GC spends 10% of time in garbage collection activities. You have to fine tune the parameters properly to achieve your pause time goals.

This related question is helpful to get some more insight into G1GC: Java 7 (JDK 7) garbage collection and documentation on G1

Answer 2

G1 is the only collection algorithm in the hotspot VM that can deal with very large heaps efficiently. However, a large heap is NOT a requirement but instead the G1 is built for situations where your application needs a very large heap. In low-heap situations, it is still outperformed by older algorithms. The same is true for the number of processors.