How to estimate the max query throughput


(Joe C) #1

My query QPS is less than 100, and was wondering what is the max possible throughput for the query.

The index has 16 shards (1 primary + 1 replica). The cluster has 11 nodes. Each node has 32 CPUs, 70G memory, 900G SSD drive. One query takes 5-10ms if I run it via kibana. If we run query sequentially, QPS is like 100.

Profiler shows that each query goes through 16 shards, so each node may need to query 2 shards.
Each node has 32 CPUs. The default search thread pool size is 3/2 * 32 = 48. So 48/2 = 24 such a query can run in parallel on a node. So the QPS with parallel query could be 4,800.

I can image given concurrency overhead or maybe some part of a query cannot be parallel. So the real QPS must be smaller than 4800.

However my QPS is 100. It is like there is no parallel at all... Is this expected?


(system) #2

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.