Performance regression on Fortunes Windows

[sebastienros]

There is a throughput regression for the Fortunes benchmarks on Windows (only).

The minimum set of commits is b4e14b4...3232ad3

NB: there is a perf improvement right before that I haven't ruled out as unrelated. This regression could be a fix to a previous issue.

Before

crank --config https://raw.githubusercontent.com/aspnet/Benchmarks/main/scenarios/database.benchmarks.yml --scenario fortunes --profile aspnet-citrine-win --application.framework net8.0 --application.aspNetCoreVersion 8.0.0-preview.5.23253.15 --application.runtimeVersion 8.0.0-preview.5.23253.15 --application.sdkVersion 8.0.100-preview.5.23254.3

After

crank --config https://raw.githubusercontent.com/aspnet/Benchmarks/main/scenarios/database.benchmarks.yml --scenario fortunes --profile aspnet-citrine-win --application.framework net8.0 --application.aspNetCoreVersion 8.0.0-preview.5.23255.6 --application.runtimeVersion 8.0.0-preview.5.23254.1 --application.sdkVersion 8.0.100-preview.5.23254.20

[sebastienros]

The improvement that we can see a few runs before is due to one of these changes:

b02d7a1...08ba40d

Before:

crank --config https://raw.githubusercontent.com/aspnet/Benchmarks/main/scenarios/database.benchmarks.yml --scenario fortunes --profile aspnet-citrine-win --application.framework net8.0 --application.aspNetCoreVersion 8.0.0-preview.5.23251.10 --application.runtimeVersion 8.0.0-preview.5.23251.7 --application.sdkVersion 8.0.100-preview.5.23228.7

After:

crank --config https://raw.githubusercontent.com/aspnet/Benchmarks/main/scenarios/database.benchmarks.yml --scenario fortunes --profile aspnet-citrine-win --application.framework net8.0 --application.aspNetCoreVersion 8.0.0-preview.5.23251.10 --application.runtimeVersion 8.0.0-preview.5.23252.1 --application.sdkVersion 8.0.100-preview.5.23228.7

[jeffschwMSFT]

Looking at the ranges, it is not obvious. Adding @tannergooding as those commits look the most interesting.

[tannergooding]

There was a bug introduced in #85536 (the improvement range) that was fixed in #85714 (the regression range). It's likely this pair of changes

[sebastienros]

Thanks for confirming.