Motivation:
Catching up on recent netty changes, simplifying data flow through channel pipeline
Modifications:
- Replaced IChannelHandlerInvoker with direct methods on IChannelHandlerContext implementation.
- Ported changes in AbstractChannel, DefaultChannelPipeline, etc. to reduce coupling through casting.
Result:
IChannelHandlerInvoker is removed, perf improved, it is easier to implement IChannel or IChannelPipeline from scratch.
Motivation:
In order to avoid buffer copies and yet perform IO efficiently,
it is essential to support sending multiple byte sequences in one call.
Also there should be no additional constraints on IByteBuffer implementations
in order to be accepted by TcpSocketChannel for sending.
Modifications:
- IoBufferCount, GetIoBuffer(), GetIoBuffers() are ported to IByteBuffer and implementations.
- ChannelOutboundBuffer is extended to leverage *IoBuffer* members in order to prepare buffer batches for sending.
- TcpSocketChannel and its ancestors were extended to support sending (sync and async) a list of array segments.
- Extra: fixes WriterIndex positioning in ByteBufferUtil.EncodeString(..).
- Extra: TLS support and exception reporting in benchmark (commented out).
- Extra: fixes async send path in TcpSocketChannel and AbstractSocketChannel.
Result:
A benchmark of single client channel writing to a single server channel runs 3 times faster at 150K op/sec (vs 53K op/sec before) with no increase in CPU consumption.
CompositeByteBuffer can be used directly in IChannel.WriteAsync.
Max Gortman