[rabbitmq-discuss] Queue replication (I know, I know)

[Matthew Sackman]

On Wed, Oct 20, 2010 at 04:03:04PM +0100, Simon MacMullen wrote:
> On 20/10/10 16:00, Derek Chen-Becker wrote:
> >I'm assuming there's some performance overhead with transacted
> >messages
> 
> Yes - basically the cost of an fsync() per commit since we need to
> know the message is on disk.

Note that multiple txns that commit at the same time can in some cases
be coalesced into a single fsync. So you might see < 1 fsync per
transaction.

Matthew