SignalR per-invocation allocations

[davidfowl]

I'm building a game that requires low latency and there's some SignalR overhead (that's pretty low tbh) but impossible to get rid of. The game I'm building is sending LOTS of messages (60 FPS) so these are now the top allocations remaining.

Some of them we can't remove, we might be able to pool objects but others we should leave alone. The ones I am interested in improving:

• System.Object[]
  • Microsoft.AspNetCore.SignalR.ClientProxyExtensions.SendAsync(Microsoft.AspNetCore.SignalR.IClientProxy, string, object, System.Threading.CancellationToken)
  • Microsoft.AspNetCore.SignalR.Protocol.JsonHubProtocol.BindTypes(ref System.Text.Json.Utf8JsonReader, System.Collections.Generic.IReadOnlyList)
• Microsoft.AspNetCore.SignalR.Internal.DefaultHubActivator\<T>
• string allocations (https://github.com/dotnet/aspnetcore/issues/41342)
• System.Collections.Concurrent.ConcurrentDictionary<,>.Enumerator (https://github.com/dotnet/runtime/issues/25448)
  • Microsoft.AspNetCore.SignalR.DefaultHubLifetimeManager.SendToAllConnections(string, System.Object[], System.Func<Microsoft.AspNetCore.SignalR.HubConnectionContext, object, bool>, object, System.Threading.CancellationToken)
• System.Byte[]
  • Microsoft.AspNetCore.Internal.MemoryBufferWriter.ToArray()
  • Microsoft.AspNetCore.SignalR.Protocol.HubProtocolExtensions.GetMessageBytes(Microsoft.AspNetCore.SignalR.Protocol.IHubProtocol, Microsoft.AspNetCore.SignalR.Protocol.HubMessage)
  • Microsoft.AspNetCore.SignalR.SerializedHubMessage.GetSerializedMessage(Microsoft.AspNetCore.SignalR.Protocol.IHubProtocol)
• Microsoft.AspNetCore.SignalR.SerializedHubMessage
• System.Collections.Generic.List\<string> - Seems like we allocate a list of empty stream ids since it's always in the payload even when it's empty. (https://github.com/dotnet/aspnetcore/pull/41344)
• There are 3 state machine boxes that allocate per operation (read/write)
  • Microsoft.AspNetCore.Server.Kestrel.Core.Internal.Http.HttpRequestStream.ReadAsyncInternal() (https://github.com/dotnet/aspnetcore/pull/41345)
  • System.IO.Pipelines.StreamPipeReader.ReadAsync.__Core|36_0() (https://github.com/dotnet/runtime/pull/68457)
  • System.Net.Security.SslStream.ReadAsyncInternal() https://github.com/dotnet/runtime/issues/68467

Type	Allocations	Bytes	Average Size (Bytes)
- System.Runtime.CompilerServices.AsyncTaskMethodBuilder\<>.AsyncStateMachineBox\<>	2,418	477,152	197.33
- System.Object[]	1,659	53,928	32.51
- Microsoft.AspNetCore.SignalR.Protocol.InvocationMessage	1,631	91,336	56
- System.String	1,629	54,420	33.41
- System.Byte[]	862	302,374	350.78
- System.Collections.Concurrent.ConcurrentDictionary\<,>.Enumerator	858	55,040	64.15
- Microsoft.AspNetCore.SignalR.SerializedHubMessage	825	72,600	88
- System.Object	825	19,800	24
- System.Int32[]	808	29,196	36.13
- System.Collections.Generic.List\<>	808	25,856	32
- System.Collections.Generic.Dictionary\<,>.Entry[]	806	96,720	120
- System.Collections.Generic.Dictionary\<,>	806	64,480	80
- BombRMan.Hubs.GameServer	806	45,136	56
- Microsoft.AspNetCore.SignalR.Protocol.CompletionMessage	806	45,136	56
- Microsoft.Extensions.DependencyInjection.ServiceLookup.ServiceProviderEngineScope	806	38,688	48
- Microsoft.AspNetCore.SignalR.Internal.DefaultHubActivator\<BombRMan.Hubs.GameServer>	806	25,792	32

[ghost]

Thanks for contacting us.

We're moving this issue to the .NET 7 Planning milestone for future evaluation / consideration. We would like to keep this around to collect more feedback, which can help us with prioritizing this work. We will re-evaluate this issue, during our next planning meeting(s). If we later determine, that the issue has no community involvement, or it's very rare and low-impact issue, we will close it - so that the team can focus on more important and high impact issues. To learn more about what to expect next and how this issue will be handled you can read more about our triage process here.

[ghost]

Thanks for contacting us.

We're moving this issue to the .NET 8 Planning milestone for future evaluation / consideration. We would like to keep this around to collect more feedback, which can help us with prioritizing this work. We will re-evaluate this issue, during our next planning meeting(s). If we later determine, that the issue has no community involvement, or it's very rare and low-impact issue, we will close it - so that the team can focus on more important and high impact issues. To learn more about what to expect next and how this issue will be handled you can read more about our triage process here.

[davidfowl]

We should be able to reduce the enumerator allocations with pooling (https://github.com/dotnet/runtime/pull/38296)