AkkaPduCodec serializer benchmark

[Aaronontheweb]

Changes

Measures the memory and throughput impact of the AkkaPduCodec - essentially simulates inbound read / write latency.

Remarks

Worth noting: understates the level of GC impact in the real Akka.Remote endpoint system - everything here gets GC0; in Akka.Remote it's GC1/2 since these messages also get delivered into the DotNetty across scope boundaries.

[Aaronontheweb]

// * Summary *

BenchmarkDotNet=v0.13.1, OS=Windows 10.0.19044.1889 (21H2)
AMD Ryzen 7 1700, 1 CPU, 16 logical and 8 physical cores
.NET SDK=6.0.201
  [Host]     : .NET 6.0.3 (6.0.322.12309), X64 RyuJIT
  DefaultJob : .NET 6.0.3 (6.0.322.12309), X64 RyuJIT

Method	Mean	Error	StdDev	Gen 0	Gen 1	Allocated
WritePayloadPdu	2,206.3 ns	5.18 ns	4.85 ns	0.3969	0.0031	1,664 B
DecodePayloadPdu	2,730.5 ns	31.49 ns	26.30 ns	0.4063	-	1,704 B
DecodePduOnly	165.4 ns	0.59 ns	0.53 ns	0.1090	-	456 B
DecodeMessageOnly	2,172.3 ns	7.19 ns	6.37 ns	0.2563	-	1,080 B
DeserializePayloadOnly	214.8 ns	2.50 ns	2.21 ns	0.0398	-	168 B

[Aaronontheweb]

Learnings here:

1. DecodeMessageAndAck is the slow path - probably all of the ActorPath parsing. Still, what's weird is that we should be using caching aggressively here (unless I designed the benchmark such that the cache never gets hit.)
2. Read-side is still ultimately the bottleneck;
3. The write-side isn't that much faster than the read-side.

[Aaronontheweb]

Think I have this fully working now.

[Aaronontheweb]

// * Summary *

BenchmarkDotNet=v0.13.1, OS=Windows 10.0.19044.1889 (21H2)
AMD Ryzen 7 1700, 1 CPU, 16 logical and 8 physical cores
.NET SDK=6.0.201
  [Host]     : .NET 6.0.3 (6.0.322.12309), X64 RyuJIT
  DefaultJob : .NET 6.0.3 (6.0.322.12309), X64 RyuJIT

Method	Mean	Error	StdDev	Gen 0	Allocated
WritePayloadPdu	2,471.4 ns	16.62 ns	14.73 ns	0.4219	1,768 B
DecodePayloadPdu	3,033.6 ns	11.12 ns	9.86 ns	0.4375	1,832 B
DecodePduOnly	226.9 ns	1.83 ns	1.53 ns	0.1145	480 B
DecodeMessageOnly	2,323.9 ns	16.67 ns	14.78 ns	0.2813	1,184 B
DeserializePayloadOnly	225.5 ns	0.84 ns	0.75 ns	0.0398	168 B

Key cache statistics on the slow path:

[Addr1] ResolveCache entries: [1]
[Addr1] PathCache entries: [4]
[Addr1] AddressCache entries: [0]
[Addr2] ResolveCache entries: [0]
[Addr2] PathCache entries: [0]
[Addr2] AddressCache entries: [0]
[Addr1] Used ResolveCache for recipient? True
[Addr1] Used PathCache for recipient? True
[Addr1] Used ResolveCache for sender? False
[Addr1] Used PathCache for sender? True
[Addr2] Used ResolveCache for recipient? False
[Addr2] Used PathCache for recipient? False
[Addr2] Used ResolveCache for sender? False
[Addr2] Used PathCache for sender? False
[Addr1] Used AddressCache for sys1? False
[Addr2] Used AddressCache for sys2? False