We have a few problems with the perf benchmarks. First they take an enormous amount of time to run, both locally and on Evergreen. The most recent build took 42 minutes. This excessive time makes it take longer to investigate performance issues.

A second issue is that some of the individual benchmarks run in a few seconds while others take 5 minutes. We should standardize this so that all the benchmarks run in a similar time frame. For example in the most recent build, TestSmallDocBulkInsert and TestGridFsDownload both took 8 seconds while TestFindOneByID took 5 minutes. This could explain why some benchmark appear less stable and have more run-to-run variance than others.

I propose we make each benchmark run for a minimum of 30 seconds and a maximum of 60.