Optimal JVM Heap size?

Elastic Stack Elasticsearch
1

Hi All,

I am looking for verification on the optimal JVM Heap size configuration for our production cluster.

We are running on physical nodes with 500 GB of RAM each.
I have read two conflicting pieces of advice regarding heap sizing:

  1. The 50% Rule: Set heap to 50% of available RAM

  2. The 31GB Limit: Never set heap above ~31GB to maintain Compressed OOPs.

My Query:
Is strictly capping the heap at 31GB the correct approach for a 500GB node, or is there any scenario where a larger heap (e.g., 64GB+) is recommended?

Thanks in advance for the help!

2

This is not an accurate summary of the relevant page in the manual (emphasis mine):

Set Xms and Xmx to no more than 50% of the total memory available to each Elasticsearch node. […] The 50% guideline is intended as a safe upper bound on the heap size. You may find that heap sizes smaller than this maximum offer better performance, for instance by allowing your operating system to use a larger filesystem cache.