Open jsimnz opened 2 years ago
For future reference - I had a play around with this in https://github.com/sourcenetwork/defradb/pull/359 taking the run-count all the way up to 40 and there was still significant differences between runs. I do think the host machine is too unstable for our benchmarks and we should look into sorting that out.
At the moment benchmarks are inconsistent in their results despite no notable code changes.
The entire goal of the benchmarks is to have a reliable metric to track the relative perf changes over time. Which is why they are run on a dedicated "clean" ec2 instance for each individual run.