We are planning to upgrade our Elasticsearch cluster to 7.X (JAVA 1.10). Is there any advice on the preferred method of garbage collection when using this particular combination, like CMS versus G1GC?
Our current hardware is 12 physical nodes running CentOS 7.6, 32 VCPU’s, and 256 GB of memory per node. We run three instances of ES per node since we max out each instances memory to 31GB per this recommendation. We control their resources via cgroups.
When moving to ES 7.x, we would like to move to larger instances of ES that utilize the G1GC. Maybe just one instance per node with 186GB of RAM it? Would this be a better way to go now that G1GC is an option? If not, what would be the recommended setup/configuration for such hardware?