Elastic from what I can tell recommends the use of RAID0, when using RAID, as data resiliency/redundancy is intended to be handled at the application layer via replicas + allocation awareness.
Split the CPU and RAM resources the your existing Elasticsearch instance uses by 4, then run 4 instances of Elasticsearch on the same node.
This can most easily be achieved by using containers (either via Docker or Kubernetes)
You could also split it by two (2) and have two (2) RAID 0 arrays with two (2) instances of Elasticsearch each using one (1). Though this would be mixing both options 1 and 2, and probably adds unneeded complication/management overhead.
If you were to provide your reasoning behind using multi-path currently, it might help in providing a more accurate/useful solution.
What I like about using multi-path is all the disk is being used and full i/o achieved.
if I loose one disk I only have 1/4 of shard unavailable untill I replace the a disk.
performance is what I am after. and seems like you have good suggestion, I didn't thought about 2nd 3rd option at all.
was only thinking how I can setup raid
each data node has 98 gig ram, and on 10 gig network.
Apache, Apache Lucene, Apache Hadoop, Hadoop, HDFS and the yellow elephant
logo are trademarks of the
Apache Software Foundation
in the United States and/or other countries.