While testing a problem with secondaries falling behind, I discovered that over time performance drops significantly on pure insert workloads. MMS screenshots attached as well as a portions of timeseries data. There are two MMS screenshots, one performance over a 3 day run, and another the shorter run performed last night.