On 02/03/2011 03:51 PM, Jeroen Geilman wrote:
On 2/3/11 8:31 PM, Joe wrote:
On 02/03/2011 11:24 AM, James R. Marcus wrote:
Hi,
I have been reading about prioritizing mail in Postfix on this
list. It seems that the answer is, there is a shared queue and
mail is not prioritized. I want to make sure that I fully
understand if this applies to my situation.
All of our servers that send mail go through our Postfix gateways. I
don't want to manage our exchange server as a relay.
I find that when we send out a lot of email and there is mail
queued, internal mail is delayed. I'm assuming the short term
solution to this is:
postqueue -s domain.com
Is there another way to prioritize internal mail delivery based on
domain?
Postfix-2.3.3-2.1.el5_2 running on CentOS 5.5 x64
One huge improvement you can make is to specify the "relay" transport
for incoming mail.
To achieve what ?
To minimize the impact of large outbound queue on incoming message
delivery latency.
That will create a separate queue from the outgoing mail which is
using the "smtp" transport,
There is one queue.
I might not have used the best term "queue" but the concept is still
correct. the queue manager implements fairness between transports, so if
incoming and outgoing mail are on two different transports it will help
a lot. Our incoming messages used to be delayed quite a bit when the
outbound queue grew to the tens of thousands. When we assigned incoming
messages to the "relay" transport, distinct from the default "smtp"
transport used for outgoing messages, we stopped seeing any problems
with inbound message delays even with huge numbers of queued outgoing
messages.
There are certainly cases for multiple postfix instances, but I wouldn't
assume that this scenario necessarily calls for such a measure. I'd try
the separate transport first, and see how it works. You may not have to
do anything else. You can always add multiple instances later if that's
really what you want to do.
There are other parameters I would look at, before running multiple
instances -
maximal_queue_lifetime, maximal_backoff_time, the process limits in
master.cf are all worth looking at.
Joe
