No difference in performance between wiredtiger and in-memory engines, any idea why?


I’ve been testing Percona Server for MongoDB with focus on in-memory engine performance.

I’ve noticed that wiredtiger engine sometimes performs better than in-memory engine for insertOne and insertMany (10000 documents).

I tried installing on host OS and on docker and with default settings for each engine. I checked the tuning articles, while insightful they didn’t offer an explanation to this weird performance of inMemory engine.

Note: I intend to use Percona Server for MongoDB for real-time persistence of inference results. The lifetime of documents won’t exceed 30 minutes. In all tests I’ve used pymongo to execute queries.

In theory when DB working set size fits into available memory there will be very small difference in performance between wiredTiger and inMemory.

You said “wiredtiger sometimes performs better”. Well “sometimes” is too vague term. Sometimes such behavior can be caused by non-ideal testing environment and sometimes by other factors.

If you are sure something is going wrong please describe steps to reproduce the issue.