Re: [hybi] hum #3: Message

[Jamie Lokier <[e-mail]>]

Dave Cridland wrote:
> On Thu Aug  5 22:50:31 2010, Greg Wilkins wrote:
> >The main down side I can think of is that it makes it harder to word
> >align header and data, but then I don't see a large demand for that.
> 
> Fixed width fields obviously allow that. Additionally, they're  
> simpler to encode and decode by a very large margin.
> 
> Finally, it means that you can build very dumb intemediaries which  
> read in the frame header in a single network read(), check the stream  
> identifier (for instance), and then hardware copy the payload between  
> interfaces. Fast reliable demultiplexing and other front-ending seems  
> like quite compelling to me.

+1.  All sensible multiplexing puts the addressing and length at the
beginning to enable cut-through forwarding without having to examine
the data itself, and allows errors to be signalled by the forwarder
during sending, in case it receives an error in the middle.

(Cut-through means forwarding data as it arrives, without waiting for
all of it first.)

If large chunks are permitted, you *must* support cut-through
forwarding otherwise the memory required for buffering is large too,
as is the forwarding latency.

-- Jamie