It turned out that we already have our own YCSB wtperf configs that represent the variants described in the original YCSB paper. They differ somewhat from how they are described in the paper (for example, they use a Pareto distribution instead of a Zipfian distribution) and also from how YCSB is measured in MongoDB (they don’t have a compressibility setting, while MongoDB uses compressibility=3). However, they are still worth measuring and reporting in our performance system.

For all the known misalignments with the paper and with MongoDB HVW we should create separate tickets to bring our configuration as close as possible to them.

This ticket is only to add these benchmarks to evergreen + report all the meaningful stats to evergreen/atlas.