I am a newbie to elastic search and working on profiling our elastic search cluster. We are using i3.4x, that means 122G of RAM and 16 vCPUs.
Currently, we are running 70G ES JVM (G1GC) and rest memory for Lucene FScache. Our application is aggregations intensive. We are noticing that our cluster is choking on compute.
I tried decreasing JVM heap to 31.9 G but it dies with OOM exception for certain queries. So my question is, should I run 2 ES JVMs on same machine with 31.9 G heap each or it won't make much difference as compared to 1 single 70G JVM?
PS - We have 20 such nodes in cluster.