Initial report is here and now I am working on making this useful to others. The Summary section in the report has the full description of the regressions. Some of the conclusions are based on vmstat and iostat data I collect via helper scripts and I hope to figure out how to show that via ftdc.
The summary is that the insert rate is between 2X to 3X faster with 4.4 prior to the durable history merge compared to 4.4 with the durable history merge. I use the names mo44pre and mo44dh (or mo44dh0315) in the report.
The workload is the insert benchmark which is open source but is not in DSI. I will provide command lines. The benchmark used 1 client thread and inserted 50M docs. The database fits in RAM, but tests were run with big-cache (database fits in WT) and small-cache (database doesn't fit in WT). The regression was much larger for small cache.
This might be related to BF-16588 and BF-16496