few minor MemoryCache perf improvements

[adamsitnik]

I just took a quick look at the source code to see if we could get some gains in the Caching TechEmpower benchmark. This PR is a free lunch, there is definitely more low hanging fruits there and I hope that it becomes one of our .NET 6 goals.

The performance repo microbenchmarks (dotnet/performance#1598) show 5 to 15% perf improvement

Method	Toolchain	Mean	Ratio
GetHit	\after\CoreRun.exe	137.61 ns	0.86
GetHit	\before\CoreRun.exe	159.17 ns	1.00
TryGetValueHit	\after\CoreRun.exe	131.57 ns	0.84
TryGetValueHit	\before\CoreRun.exe	157.07 ns	1.00
GetMiss	\after\CoreRun.exe	98.56 ns	0.87
GetMiss	\before\CoreRun.exe	113.47 ns	1.00
TryGetValueMiss	\after\CoreRun.exe	97.62 ns	0.86
TryGetValueMiss	\before\CoreRun.exe	113.05 ns	1.00
SetOverride	\after\CoreRun.exe	362.80 ns	0.95
SetOverride	\before\CoreRun.exe	381.01 ns	1.00

The TechEmpower benchmark shows +6k RPS (from 234,210 240,778) to which translates to +2.75% RPS

[ghost]

Tagging subscribers to this area: @eerhardt, @maryamariyan
See info in area-owners.md if you want to be subscribed.

Issue Details

Description:	I just took a quick look at the source code to see if we could get some gains in the Caching TechEmpower benchmark. This PR is a free lunch, there is definitely more low hanging fruits there and I hope that it becomes one of our .NET 6 goals. The performance repo microbenchmarks (dotnet/performance#1598) show 5 to 15% perf improvement Method Toolchain Mean Ratio GetHit \after\CoreRun.exe 137.61 ns 0.86 GetHit \before\CoreRun.exe 159.17 ns 1.00 TryGetValueHit \after\CoreRun.exe 131.57 ns 0.84 TryGetValueHit \before\CoreRun.exe 157.07 ns 1.00 GetMiss \after\CoreRun.exe 98.56 ns 0.87 GetMiss \before\CoreRun.exe 113.47 ns 1.00 TryGetValueMiss \after\CoreRun.exe 97.62 ns 0.86 TryGetValueMiss \before\CoreRun.exe 113.05 ns 1.00 SetOverride \after\CoreRun.exe 362.80 ns 0.95 SetOverride \before\CoreRun.exe 381.01 ns 1.00 The TechEmpower benchmark shows +6k RPS (from 234,210 240,778) to which translates to +2.75% RPS
Author:	adamsitnik
Assignees:	-
Labels:	area-Extensions-Caching
Milestone:	-

[stephentoub]

This isn't inlined otherwise? Also, ScheduleTask looks pretty small; it's not inlined automatically (which would defeat your goal here)?

[adamsitnik]

before my changes, the StartScanForExpiredItems was not getting inlined:

now StartScanForExpiredItemsIfNeeded does get inlined an ScheduleTask does not:

I was simply not sure if _options.ExpirationScanFrequency < utcNow - _lastExpirationScan won't get translated to too much IL and just to be sure I've applied the attribute

[stephentoub]

I was simply not sure if _options.ExpirationScanFrequency < utcNow - _lastExpirationScan won't get translated to too much IL and just to be sure I've applied the attribute

Ok, but the same argument goes for whether ScheduleTask may get inlined automatically even though you don't want it to.

[adamsitnik]

but the same argument goes for whether ScheduleTask may get inlined automatically even though you don't want it to

It's quite big (from the IL perspective) and I would not expect it to get inlined. Do you want me to ensure it by using an attribute?

[kunalspathak]

Perf lab data showing improvements:

https://pvscmdupload.blob.core.windows.net/reports/allTestHistory/refs/heads/master_x64_Windows%2010.0.18362/Microsoft.Extensions.Caching.Memory.Tests.MemoryCacheTests.GetHit.html

https://pvscmdupload.blob.core.windows.net/reports/allTestHistory/refs/heads/master_x64_Windows%2010.0.18362/Microsoft.Extensions.Caching.Memory.Tests.MemoryCacheTests.GetMiss.html

[kunalspathak]

Related improvement link: DrewScoggins/performance-2#3516