[rabbitmq-discuss] Schedule Messages

[David Wragg]

Matthew Sackman <matthew at rabbitmq.com> writes:
> On Fri, Oct 01, 2010 at 01:45:46PM +0100, Michael Bridgen wrote:
>> In the case of plugins, using a direct connection is effectively the
>> same thing, by the way.
>
> No it's not. The msg may be fully on disk and not in RAM, at which point
> you have to issue a disk read regardless of whether the client is
> embedded or not.
>
> Furthermore, we do not store the message headers separately from the msg
> body. Thus the only thing you're saving by getting one, not the other,
> is network transfer. And given the average message body size, ethernet
> frame size and the fact that we turn nagel off, I'm not convinced in the
> average case you'd actually save anything at all, but I could be missing
> something.

We'd get the network traffic saving immediately.  And if worthwhile,
later on we could store message bodies separately, above a certain size
threshold.  At the moment there is no motivation to consider doing that,
because AMQP always transmits the properties and body together.

And I doubt there is a meaningful notion of the average message body
size.  Different AMQP applications have different characteristics.

-- 
David Wragg
Staff Engineer, RabbitMQ
SpringSource, a division of VMware