Searchable Snapshot and HDD vs SSD/NVMe Performance

Elastic Stack Elasticsearch
1

Hi All,

I have a question that I couldn't really find an answer to:

When using searchable snapshots, is the data stored and read in a "sequential" order, in a way that would be ideal for using HDDs with, or is the "recommendation" to use SSDs/NVMe for searchable snapshot nodes.

For context, the nodes would be AWS EC2 instances mostly of the "storage optimized" class.

If there are benchmarks somewhere of a comparison of these 2 storage mediums somewhere as well, that would be interesting.

2

The data on disk on the node itself is laid out pretty much just like any other Lucene data, with the usual expectations about access patterns: multiple concurrent readers, each of which will often only read forwards but may skip rather than doing purely sequential reads, and others will do more of a random pattern. The data is retrieved from snapshot in a fairly chunky sequential fashion, but again with multiple concurrent writers that probably won't play nicely with spinning disks.

TLDR yes use SSDs.

1 Like
3

Great! Thanks for the info :slight_smile: