Consider custom implementation of FrameWriter for sockets.

[cbaxter]

Sockets implement custom send that accepts collection of buffers; thus would no longer require Buffer.BlockCopy to build up single message buffer. Determine impact on performance and act accordingly.

[cbaxter]

1,000,000 messages of 4,096 bytes each achieves the following metrics:

PUB/SUB Channel

• Message Latency = 3.08 [us]
• Message Throughput = 161904.85 [msg/s]
• Data Throughput = 5305.30 [Mb/s]

REQ/REP Channel

• Message Latency = 15.17 [us]
• Message Throughput = 32962.33 [msg/s]
• Data Throughput = 1080.11 [Mb/s]

Not sure if there will be a huge gain from creating custom implementations given the Buffer.BlockCopy is adding negligible overhead according to ANTS Performance Profiler. Thoughts?

[cbaxter]

Spiked out custom SocketBlockingFrameWriter implementation against REQ/REP channel. Throughput actually dropped by ~2,000 msg/sec. Buffer.BlockCopy is already optimized to move around buffers very efficiently (basically doing an unsafe copy). Looked in to Socket.Send implementation and is pinning each individual buffer prior to calling:

UnsafeNclNativeMethods.OSSOCK.WSASend_Blocking(this.m_Handle.DangerousGetHandle(), buffersArray, count, out num2, socketFlags, IntPtr.Zero, IntPtr.Zero);

Not worth the added code or complexity required for implementation.

NOTE: GitHub not allowing selection of 'Rejected' tag.