Since many map and reduce tasks run in parallel, Informatica recommends setting the max heap size based on the system memory to reduce or avoid swapping tasks.

For a parallel garbage collector, Informatica recommends setting the -XX:ParallelGCThreads property to a smaller value such as 2 to reduce context switching overhead since many map and reduce tasks are executed by a task tracker.

For example, assume that a slave node has 32GB RAM, and it is configured to have 32 total slots including map slots and reduce slots. In this case, the JVM maximum heap size value '-Xmx' can be specified up to a maximum of 1GB.

However, this does not rule out swapping. Other operating system processes can compete for memory and the DTM process running on the Hadoop node consumes memory outside the JVM heap. The default value of '-Xmx512M' is good enough for most of the cases. However, Informatica recommends setting the correct value by experimenting with different values to ensure swapping does not impact performance.