This is part of The Pile, a partial archive of some open source mailing lists and newsgroups.
To: Aaron Lehmann <aaronl@vitelus.com> Subject: Re: [svlug] exim? Date: Sat, 10 Feb 2001 09:32:32 -0800 From: J C Lawrence <claw@kanga.nu> On Sat, 10 Feb 2001 00:13:22 -0800 Aaron Lehmann <aaronl@vitelus.com> wrote: > On Sat, Feb 10, 2001 at 12:10:59AM -0800, J C Lawrence wrote: >> I'm not aware of any. The tests I did for a client a few months >> ago were not rigorous enough (or normalised) for useful >> publication (ie it would have been easy to poke holes and ignore >> the revealed trends). > Well, what did you find? That Exim and Postfix had very similar performance rates at the following two load points: -- Continutally busy -- Saturated However the curves that each took to reach those load points were significantly different, with the most general characteristic being that Exim tended to be kinder on the localhost in terms of load and responsetimes at the expense of tending to maintain a more continual low level queue activity, whereas Postfix tended toward system hammering burstiness in very short-termed responses to load. > How did you perform the tests? Two to three machines (what I had on hand). One MTA, the others generating and receiving messages. Some messages generated locally to the MTA via direct injection into the spool. The stats I tracked were rate of deliveries and system load. === Date: Sat, 10 Feb 2001 10:04:14 -0800 From: Karen Shaeffer <shaeffer@best.com> To: svlug@svlug.org Subject: Re: [svlug] exim? On Sat, Feb 10, 2001 at 09:32:32AM -0800, J C Lawrence wrote: > However the curves that each took to reach those load points were > significantly different, with the most general characteristic being > that Exim tended to be kinder on the localhost in terms of load and > responsetimes at the expense of tending to maintain a more continual > low level queue activity, whereas Postfix tended toward system > hammering burstiness in very short-termed responses to load. > > > How did you perform the tests? > > Two to three machines (what I had on hand). One MTA, the others > generating and receiving messages. Some messages generated locally > to the MTA via direct injection into the spool. The stats I tracked > were rate of deliveries and system load. I presume you removed other variables from the test setup, i.e., there was a minimal and consistent set of other processes running. Did you happen to look at the code? I would guess Postfix is threaded and Exim probably isn't. === Date: Sat, 10 Feb 2001 16:27:27 -0800 From: Drew Bloechl <drew@cesspool.net> To: svlug@svlug.org Subject: Re: [svlug] exim? On Sat, Feb 10, 2001 at 03:21:38PM -0800, Rick Moen wrote: > Hmm. Several threads on the postfix-users mailing list suggest (but not > absolutely conclusively) that Postfix remains single-threaded. The > Postfix mirror sites' HTML doesn't specify. You are correct in that it does not thread. It uses IPC to communicate between the daemons that implement the various functions. === To: Karen Shaeffer <shaeffer@best.com> Subject: Re: [svlug] exim? Date: Sat, 10 Feb 2001 15:56:50 -0800 From: J C Lawrence <claw@kanga.nu> On Sat, 10 Feb 2001 10:04:14 -0800 Karen Shaeffer <shaeffer@best.com> wrote: > I presume you removed other variables from the test setup, i.e., > there was a minimal and consistent set of other processes > running. Of course -- this was a bench setup. > Did you happen to look at the code? I would guess Postfix is > threaded and Exim probably isn't. Neither is threaded. Exim is monolithic. Postfix is a well divided set of mutually untrusting executables. ===