CrateDB cpu vs ram vs more nodes

Question

I'm trying to setup CrateDB in Google Cloud for analytics through Metabase, availability is not important, the data can be reloaded, just query speed. Largest table is 50 million rows 40 columns. All tables are denormalized.

What is more beneficial for CrateDB query performance, number of nodes, CPU quantity or RAM amount?

• 6 nodes x 1 VCPU 3.75GB RAM
• 3 nodes x 2 VCPU 7.5GB RAM
• 3 nodes x 1 VCPU 15GB RAM
• 3 nodes x 4 VCPU 4GB RAM
• 1 node x 6 VCPU 22.5GB RAM

Is it better to try adding as much CPU as possible, as much RAM as possible or a balance of both?

Answer 1

it depends on your use case, but usually you go for a mixture. but what you described, i'd go with: 3 nodes x 4 VCPU 4GB RAM

cratedb is distributed by nature so you need to run it in a cluster to use its benefits.

• if you need to do load a lot of data or use memory intense queries like joins, subselects, distincts you need to have more memory.
• if you need to handle a lot of connections or do operations where a lot of shards are involved you need more cpus.
• if you do a lot of ingest you need to have fast discs and more cpus.

if you have the possibility, use ssds. spinning discs slows cratedb down a lot.