1 |
On Thu, 2005-19-05 at 09:45 -0500, kashani wrote: |
2 |
|
3 |
> Email throughput is highly dependent on how you're generating the |
4 |
> emails, RAM, disk I/O, number of emails sitting in your queue, how |
5 |
> you've ordered your emails, whether bounce handling is local or on |
6 |
> another server, spam filtering if any, and your server software. Even |
7 |
> the speed of your DNS server can start to play a part. |
8 |
|
9 |
This is all astute commentary. I agree with all of it. Fine tuning an |
10 |
email system (or any production platform) takes ... finesse. It's not |
11 |
about "installing this software or that software", but about |
12 |
understanding the behaviour of your system as a whole, and then tuning |
13 |
the performance from a position of knowledge, not from one of |
14 |
assumption. |
15 |
|
16 |
> 30/sec puts you a bit above 2.5 million a day which is probably the top |
17 |
> end of what you might see in the real world. |
18 |
|
19 |
Certainly seems a reasonable score. |
20 |
|
21 |
I've run 3-5 million emails a day through a platform based on Qmail. |
22 |
Qmail has architectural bottlenecks on how it processes new messages; |
23 |
likewise you need to take care to shunt traffic to misbehaving domains |
24 |
to alternate queues - but making it sing was little to do with Qmail |
25 |
itself and very much to do with knowing what the group of systems as a |
26 |
whole were doing and making informed judgements thereon. |
27 |
|
28 |
AfC |
29 |
Sydney |
30 |
|
31 |
-- |
32 |
Andrew Frederick Cowie |
33 |
|
34 |
Technology strategy, managing change, establishing procedures, |
35 |
and executing successful upgrades to mission critical business |
36 |
infrastructure. |
37 |
|
38 |
http://www.operationaldynamics.com/ |
39 |
|
40 |
Sydney New York Toronto London |
41 |
|
42 |
-- |
43 |
gentoo-server@g.o mailing list |