Tuning Java* Garbage Collection for Spark* Applications

Spark is gaining wide industry adoption due to its superior performance, simple interfaces, and a rich library for analysis and calculation. Like many projects in the big data ecosystem, Spark runs on the Java Virtual Machine (JVM). Because Spark can store large amounts of data in memory, it has a major reliance on Java’s memory management and garbage collection (GC). New initiatives like Project Tungsten will simplify and optimize memory management in future Spark versions. But today, users who understand Java’s GC options and parameters can tune them to eek out the best the performance of their Spark applications. This article describes how to configure the JVM’s garbage collector for Spark, and gives actual use cases that explain how to tune GC in order to improve Spark’s performance. We look at key considerations when tuning GC, such as collection throughput and latency.

Download complete article PDF