Date: Fri, 23 Sep 2005 11:56:37 -0500 (CDT) From: Randy Schultz <schultz@sgi.com> To: Deepak Naidu <deepak_nai@yahoo.com> Cc: freebsd-advocacy@freebsd.org, freebsd-questions@freebsd.org Subject: Re: Performance of mailserevr in FreeBSD 5.4 Message-ID: <Pine.BSF.4.63.0509201604080.65257@tdream.americas.sgi.com> In-Reply-To: <20050920174443.78027.qmail@web34614.mail.mud.yahoo.com> References: <20050920174443.78027.qmail@web34614.mail.mud.yahoo.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, 20 Sep 2005, Deepak Naidu spaketh thusly:
-}Thanx Randy,
-}
-} It would be good, if I have some data of posted doc regarding this... or of
-}your own experience. Thanx for your advise ....
Ok. I'm posting this to advocacy as well in case any find it useful or
at least interesting.
Here's a summation of the testing I did and the results of those tests.
As we were only looking for a few things this is far from a scientific
analysis so FWIW.
Test background/setup:
I wanted to test various configs to see what gave the best throughput in
a relay configuration for our real-time requirements, i.e. ave # of
recipients, ave size of each email, etc.
- OS's: FreeBSD 5.4 with and w/o softupdates, Fedora Core 3, Fedora Core 4.
The FC installations used ext3 for the filesystems. I wanted to try
with Reiser but ran out of time.
- Sink systems simply threw everything to /dev/null. We had multiple sink
systems to ensure they were never a chokepoint.
- Source systems used fbsd's postal package as:
postal -m 11 -p 4 -c 2 d3 usernames -
where usernames contained 2 and 3 usernames. No significant
difference was found using 2 or 3 usernames.
- Test relay was a dual-proc 800 MHz system PC with 1 GB RAM and the
entire system on 1 5400 RPM drive. (I know - nobody would ever run
a mail server with /var/spool on the root drive but we were really
just looking for difference percentages, not max throughput and had
some old systems lying around... ;)
- All installs were default installs. I thought about tweaking this or
that, e.g. postfix has some notes on things to do for high-volume
installations that we didn't do. I wanted out-of-the-box as much as
possible. Tweaking can quickly turn into a slippery slope of just 1
more here and 1 more there. I figured we can tweak all we want for
more specific needs as they arise.
- The versions of sendmail and postfix were whatever was current stable
in June, compiled locally with default build instructions.
So, given all that, here's the #'s I came up with.
Sendmail
Linux: 47,000 emails/hr
FreeBSD: 66,000 emails/hr
this is about a 40% increase in throughput.
Postfix
Linux: 86,000 emails/hr
FreeBSD: 223,000 emails/hr
this is about a 260% increase in throughput.
The above data is for FC4 and FBSD 5.4 with softupdates enabled. My
apologies but I can't find the notes for FC3 and FBSD w/o softupdates. FC4
was faster but not by much, and by about the same % for sendmail and postfix.
Softupdates showed a similar pattern(as expected) - not as fast as with
softupdates. IIRC the data for postfix/fbsd w/o softupdates was still
around 190k emails/hr(this is from memory so take it with a grain of salt).
I was amazed at the difference with postfix. Whatever Mr. Venema did inside
postfix really works well on fbsd. Even though I can't find the #'s I
remember my amazement that fbsd w/o softupdates still smoked FC4 when using
postfix.
Pls remember this is far from a scientific analysis. Your mileage will vary
in many ways. I would recommend doing your own tests with your own criteria.
--
Randy (schultz@sgi.com) 715-726-2832 email bodhisattva <*>
"There is no fire like passion, there is no shark like hatred,
there is no snare like folly, there is no torrent like greed."
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.4.63.0509201604080.65257>