Currently, in our cluster, this is the configuration,
Nodes - 7 nodes(3master, 4datanodes, write happens 1 month once, so mostly read cluster)
ES - 6.4
Memory - 128 GB(ES - 32 GB)
Cores - 16
Shards - 130+ (each shard 30-40 GB)
Indices - 5
During peak hours we're getting search rejections due to the QPS, so we're planning to add more nodes to the cluster. W.r.t shall we divide our current cluster nodes from (128GB, 16C) to (64GB, 8C) & still 32 GB will be allocated to ES, does this help in more parallelism in processing requests?
Which is better out of these two, less nodes with more memory & CPU vs more nodes with less memory & CPU for Search processing from a scalability perspective?