Re: [hybi] WS & Nagle in use

[Patrick McManus <[e-mail]>]

On Mon, 2011-03-21 at 08:19 -0500, Greg Longtin wrote:

> Hence, should the specs state clearly that if a server or client WS
> application receives a 'multiple message' Nagle packet, that all messages
> should be decoded from it?
> 

TCP is a byte stream protocol - a single WS message might span multiple
tcp packets with just 1 data byte in each, or it might be carried with
thousands of other WS messages in a single TCP segment which itself is
fragmented across multiple IP packets.  Or of course there might be 1
websocket message per TCP segment.

Websockets cannot restrict those options in any way, and receivers must
deal with it (firefox included, because you asked.). The TCP API is
bytestream based - it doesn't give reliable insight into how the data
was arranged when it arrived.

That doesn't really have much to do with nagle. You might see coalescing
due to TCP congestion control delays or network driver scheduling delays
or any number of other various indeterminate things.

> In my testing, I'm not sure if the current clients did so.  Should I test
> the Minefield and Silverlight implementations, or are we certain that they
> do so?
>