avoid allocating SocketAddress GCHandle for every ReceiveFromAsync

[wfurt]

fixes #86513
contributes to #30797

Because we always serialize EndPoint to SocketAddress this check is never true:

runtime/src/libraries/System.Net.Sockets/src/System/Net/Sockets/SocketAsyncEventArgs.Windows.cs

Lines 863 to 864 in f60757a

	// Check if already pinned.
	if (_pinnedSocketAddress == _socketAddress)

So we and up creating GCHandle for every packet making all the logic useless.
The fix I discussed with @stephentoub is to use NativeMemory that is quite fast and same memory chunk can persist across many recipes regardless of the actually SocketAddress.
As downside we need to copy the actual socket address bytes - but it is small so it should be fast and it does not got memory with movable blocks.

Perftests do not really show much because they do not track the GCHandles.

Method	Job	Toolchain	Mean	Error	StdDev	Median	Min	Max	Ratio	RatioSD	Allocated	Alloc Ratio
ConnectAcceptAsync	Job-SWLAAA	\8.0.0\corerun.exe	4.238 ms	1.2878 ms	1.4830 ms	3.928 ms	1.816 ms	7.023 ms	1.00	0.00	1343 B	1.00
ConnectAcceptAsync	Job-DYCDTR	\main\corerun.exe	3.341 ms	0.7778 ms	0.8646 ms	3.351 ms	1.827 ms	5.361 ms	0.88	0.35	1368 B	1.02
SendAsyncThenReceiveAsync_Task	Job-SWLAAA	\8.0.0\corerun.exe	255.208 ms	15.2474 ms	17.5590 ms	256.833 ms	215.582 ms	280.909 ms	1.00	0.00	-	NA
SendAsyncThenReceiveAsync_Task	Job-DYCDTR	\main\corerun.exe	250.302 ms	10.1277 ms	10.8365 ms	251.113 ms	214.067 ms	268.417 ms	0.98	0.08	-	NA
ReceiveAsyncThenSendAsync_Task	Job-SWLAAA	\8.0.0\corerun.exe	2,540.331 ms	471.9609 ms	543.5108 ms	2,570.356 ms	1,353.512 ms	3,550.914 ms	1.00	0.00	-	NA
ReceiveAsyncThenSendAsync_Task	Job-DYCDTR	\main\corerun.exe	2,179.767 ms	330.8853 ms	381.0480 ms	2,151.268 ms	1,419.066 ms	2,880.350 ms	0.91	0.31	-	NA
SendAsyncThenReceiveAsync_SocketAsyncEventArgs	Job-SWLAAA	\8.0.0\corerun.exe	204.784 ms	41.5151 ms	47.8089 ms	220.779 ms	142.865 ms	279.819 ms	1.00	0.00	-	NA
SendAsyncThenReceiveAsync_SocketAsyncEventArgs	Job-DYCDTR	\main\corerun.exe	177.627 ms	29.7090 ms	33.0214 ms	191.337 ms	96.820 ms	208.411 ms	0.87	0.17	-	NA
ReceiveAsyncThenSendAsync_SocketAsyncEventArgs	Job-SWLAAA	\8.0.0\corerun.exe	1,899.842 ms	463.4884 ms	533.7539 ms	1,925.979 ms	636.953 ms	2,709.015 ms	1.00	0.00	-	NA
ReceiveAsyncThenSendAsync_SocketAsyncEventArgs	Job-DYCDTR	\main\corerun.exe	2,335.415 ms	340.3239 ms	391.9175 ms	2,393.339 ms	1,767.759 ms	3,198.932 ms	1.43	0.85	-	NA
ReceiveFromAsyncThenSendToAsync_Task	Job-SWLAAA	\8.0.0\corerun.exe	3,382.587 ms	868.8274 ms	1,000.5430 ms	3,373.508 ms	1,219.412 ms	4,980.546 ms	1.00	0.00	5761064 B	1.00
ReceiveFromAsyncThenSendToAsync_Task	Job-DYCDTR	\main\corerun.exe	5,021.543 ms	272.1956 ms	313.4609 ms	5,057.454 ms	3,903.799 ms	5,358.622 ms	1.69	0.83	6241688 B	1.08
SendToThenReceiveFrom	Job-SWLAAA	\8.0.0\corerun.exe	368.134 ms	15.5442 ms	17.9007 ms	367.633 ms	319.479 ms	390.570 ms	1.00	0.00	5760784 B	1.00
SendToThenReceiveFrom	Job-DYCDTR	\main\corerun.exe	341.450 ms	34.0893 ms	39.2573 ms	352.988 ms	231.940 ms	386.746 ms	0.93	0.10	6240784 B	1.08

[ghost]

Tagging subscribers to this area: @dotnet/ncl
See info in area-owners.md if you want to be subscribed.

Issue Details

---

fixes #86513
contributes to #30797

Because we always serialize EndPoint to SocketAddress this check is never true:

runtime/src/libraries/System.Net.Sockets/src/System/Net/Sockets/SocketAsyncEventArgs.Windows.cs

Lines 863 to 864 in f60757a

	// Check if already pinned.
	if (_pinnedSocketAddress == _socketAddress)

So we and up creating GCHandle for every packet making all the logic useless.
The fix I discussed with @stephentoub is to use NativeMemory that is quite fast and same memory chunk can persist across many recipes regardless of the actually SocketAddress.
As downside we need to copy the actual socket address bytes - but it is small so it should be fast and it does not got memory with movable blocks.

Perftests do not really show much because they do not track the GCHandles.

Method	Job	Toolchain	Mean	Error	StdDev	Median	Min	Max	Ratio	RatioSD	Allocated	Alloc Ratio
ConnectAcceptAsync	Job-SWLAAA	\8.0.0\corerun.exe	4.238 ms	1.2878 ms	1.4830 ms	3.928 ms	1.816 ms	7.023 ms	1.00	0.00	1343 B	1.00
ConnectAcceptAsync	Job-DYCDTR	\main\corerun.exe	3.341 ms	0.7778 ms	0.8646 ms	3.351 ms	1.827 ms	5.361 ms	0.88	0.35	1368 B	1.02
SendAsyncThenReceiveAsync_Task	Job-SWLAAA	\8.0.0\corerun.exe	255.208 ms	15.2474 ms	17.5590 ms	256.833 ms	215.582 ms	280.909 ms	1.00	0.00	-	NA
SendAsyncThenReceiveAsync_Task	Job-DYCDTR	\main\corerun.exe	250.302 ms	10.1277 ms	10.8365 ms	251.113 ms	214.067 ms	268.417 ms	0.98	0.08	-	NA
ReceiveAsyncThenSendAsync_Task	Job-SWLAAA	\8.0.0\corerun.exe	2,540.331 ms	471.9609 ms	543.5108 ms	2,570.356 ms	1,353.512 ms	3,550.914 ms	1.00	0.00	-	NA
ReceiveAsyncThenSendAsync_Task	Job-DYCDTR	\main\corerun.exe	2,179.767 ms	330.8853 ms	381.0480 ms	2,151.268 ms	1,419.066 ms	2,880.350 ms	0.91	0.31	-	NA
SendAsyncThenReceiveAsync_SocketAsyncEventArgs	Job-SWLAAA	\8.0.0\corerun.exe	204.784 ms	41.5151 ms	47.8089 ms	220.779 ms	142.865 ms	279.819 ms	1.00	0.00	-	NA
SendAsyncThenReceiveAsync_SocketAsyncEventArgs	Job-DYCDTR	\main\corerun.exe	177.627 ms	29.7090 ms	33.0214 ms	191.337 ms	96.820 ms	208.411 ms	0.87	0.17	-	NA
ReceiveAsyncThenSendAsync_SocketAsyncEventArgs	Job-SWLAAA	\8.0.0\corerun.exe	1,899.842 ms	463.4884 ms	533.7539 ms	1,925.979 ms	636.953 ms	2,709.015 ms	1.00	0.00	-	NA
ReceiveAsyncThenSendAsync_SocketAsyncEventArgs	Job-DYCDTR	\main\corerun.exe	2,335.415 ms	340.3239 ms	391.9175 ms	2,393.339 ms	1,767.759 ms	3,198.932 ms	1.43	0.85	-	NA
ReceiveFromAsyncThenSendToAsync_Task	Job-SWLAAA	\8.0.0\corerun.exe	3,382.587 ms	868.8274 ms	1,000.5430 ms	3,373.508 ms	1,219.412 ms	4,980.546 ms	1.00	0.00	5761064 B	1.00
ReceiveFromAsyncThenSendToAsync_Task	Job-DYCDTR	\main\corerun.exe	5,021.543 ms	272.1956 ms	313.4609 ms	5,057.454 ms	3,903.799 ms	5,358.622 ms	1.69	0.83	6241688 B	1.08
SendToThenReceiveFrom	Job-SWLAAA	\8.0.0\corerun.exe	368.134 ms	15.5442 ms	17.9007 ms	367.633 ms	319.479 ms	390.570 ms	1.00	0.00	5760784 B	1.00
SendToThenReceiveFrom	Job-DYCDTR	\main\corerun.exe	341.450 ms	34.0893 ms	39.2573 ms	352.988 ms	231.940 ms	386.746 ms	0.93	0.10	6240784 B	1.08

Author:	wfurt
Assignees:	wfurt
Labels:	area-System.Net.Sockets, os-windows
Milestone:	8.0.0

[wfurt]

I updated names and comments @stephentoub

[stephentoub]

LGTM

Did you happen to measure any impact to just allocating and freeing on each operation? Would that clean up any code / make the SAEA smaller?

[wfurt]

Did you happen to measure any impact to just allocating and freeing on each operation? Would that clean up any code / make the SAEA smaller?

no, but it would be easy to try. My guess is that the impact would be small - in perf tests even voiding GCHandle was small -> most benefit come from memory management IMHO