Skip to content

Regression in Caching TechEmpower benchmark #52640

Description

@adamsitnik

I wanted to provide some nice benchmark results for @maryamariyan who is currently writing the blog post about Microsoft.Extensions* improvements we did for .NET 6.

To my surprise, the CachingPlatform-intel-win has regressed from 290k+ to 270k RPS. CachingPlatform-intel (which is running on Linux) regressed from 270k+ to 255k RPS.

The regression happened on the 12th of April:

obraz

Power BI

I have taken a brief look at the profiles and it seems that it has been caused by #50778 which got also confirmed by the perf bot in DrewScoggins/performance-2#5151

The Caching benchmark is very specific, as due to caching the values are fetched from the database only once, and most of the time (40%+) is spent in... JSON serialization ;)

The JSON serialization logic can be found here:

https://github.com/aspnet/Benchmarks/blob/6e7bff20860c92fd22b511bfd160f6f153e7e96e/src/BenchmarksApps/Kestrel/PlatformBenchmarks/BenchmarkApplication.MultipleQueries.cs#L17-L39

In the case of this particular benchmark, we are serializing an array of one hundred instances of World, which is a very small class:

https://github.com/aspnet/Benchmarks/blob/6e7bff20860c92fd22b511bfd160f6f153e7e96e/src/BenchmarksApps/Kestrel/PlatformBenchmarks/Data/CachedWorld.cs

@eiriktsarpalis @layomia @steveharter is there any chance that you could take a look?

edit: struct -> class

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Type

No type

Projects

No projects

    Milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions