performance testing new server, results

[Chris Yuzik]

am.lists wrote:
> Benchmarking performance like this is so difficult to get right. There
> are so many variables. For one, you've hobbled all of the things that
> the software can do to attain better performance. I sorta agree/sorta
> disagree here. On one hand, it gives you total worst-case-scenario
> numbers, but then on the other hand, you hope your server never has
> those conditions to work under. (So you really don't know how well it
> can perform, you only know how poorly it can perform.)
Right. I wanted to see worst-case-scenario and make sure box didn't 
hang, crash, or otherwise implode. As I write this the queue is now 
empty. Total time to process the 4500 spams/hams + all the mail that hit 
our production server during the test, about 45 minutes. The current 
production server seems to average between 8k to 10k per day.

My next test will be to reduce the fuzzyocr setting down to where it's 
supposed to be, so it isn't called all the time, and to enable 
spamassassin caching.
> Here's a better question: What throughput numbers (Msgs/minute) do you
> need to attain?
Well, as I mentioned above, we're currently handling about 10k messages 
per day, but that's increasing on a monthly basis. Also, the email 
volume we send/receive isn't constant, with peak time typically between 
11am and 2pm each weekday.

I'd like to be confident that this server can handle 50k to 75k messages 
per day, including not getting overloaded during peak times.
> Under load, how did the box perform? Was it CPU-bound?
> Memory-bound/swapping? Or was it pushing a limit on network bandwidth?
> (I would assume you were NOT sending mail out during the test...)
The CPU seemed to bounce around between 34% and 80%, spending most of 
its time hovering around 50%.

The swap remained at 0k used throughout the test.

This box was not sending out any mail to the internet during the test.

Thanks for your help.

Chris