There are two issues involved in this question...
1) Cluster is non-heterogenous - Different CPUS and drive configurations.
2) Cluster has a mix of spinning rust , SSDs and newer tech. (NvME, PCIe Flash, etc...)
Can we take advantage of the faster storage for things like MapRDB/HBase or disk cache for an in-memory index like SOLR/Lucene/ES or something similar?
I know we could put the different drives in to different volumes, but has anyone tried to tier the storage?
Also what about hardware based improvements like GPUs, or other hardware accelerators?