[rabbitmq-discuss] Exchange matching performance

[Matthias Radestock]

Matthew Sackman wrote:
> On Mon, Jun 15, 2009 at 12:07:32PM -0500, tsuraan wrote:
>> I seem to remember that direct exchanges dispatch much faster than
>> topic due to the complexity of the matching in the topic exchange.
>> Now that the headers exchange is out (or at least appears to be in the
>> disk paging branch), how does that perform?  It seeks like it would be
>> similar to the direct exchange, but with multiple possible keys.  I
>> haven't done any speed testing yet; I'm just wondering what the
>> expectation would be.
> 
> I can't comment as I didn't write that code, and the person who did is
> currently on holiday. Beware that header exchange is defined in AMQP
> 0-9-1 and that what you have checked out is in fact an interesting mix
> of 0-8 and 0-9-1. Clients almost certainly will break.

The introduction of headers exchanges does not change the protocol's 
wire format or require any additional AMQP commands. So clients are 
unaffected.

As for efficiency; the current implementation of headers exchanges 
performs about the same as topic exchanges.


Matthias.