I'm building ES 2.2.x cluster with
- 5 dedicated data nodes, 8 cores, 20Gb RAM each; handling 1500Gb of data (about 150 shards total).
- 2 dedicated client nodes, 4 cores, 12Gb RAM each.
- 3 dedicated master nodes, 2 cores, 2Gb RAM each.
I have two questions regarding this:
- What is heap/RAM size ratio for master and client nodes? Since they don't use any disk cache, can I set rather high to about 80%? (My queries are mostly aggregations)
- Would 2 cores / 2GB RAM for each master node be enough for my setup?