On Mon, 1 Oct 2001 15:22:40 -0400 Jim Kutter <[EMAIL PROTECTED]> wrote:
> Also can anyone give me hard times for qmail+mailman with a very > large list (> 30K members)? I've heard it's "really fast" but that > doesn't help me make the sale for qmail... Hard stats in the MTA field are difficult if not impossible to achieve. There are just too many external factors such as variant RTT, percentage and distribution of slow MXes, distribution of target MXes and their percentage distribution over the load, size of RCPT TO bundling, etc etc etc. Statistically, its a large and thorny problem with any particular numbers derived in one instance likely not applicable elsewhere except as a hint. The most fundamental aspect of MTAs is that they are really not limited by MTA software performance. They are limited by disk IO. While there are minor differences in the exact behaviour of the different MTAs when they run up against the disk IO wall, the numbers when they all do are remarkably similar. This shouldn't be surpising when you consider the requirements for commited writes via open()/fsync()/close(). There are three performant MTAs currently on the Open Source market: Exim, Postfix, and QMail. Exim is a monolothic design much in the manner of SMail, from which it was derived. The author, Philip Hazel, is responsive and helpful. It has excellent documentation, an active support community and the most human readable and understandable config files of any MTA I've seen. it also happens to come with an excellent Mailman HOWTO. In tests here Exim's performance curves were similar to Postfix's, with a lower attack curve and similar delivery rates at the inflection points of constantly-busy and queue-saturated. Exim is particularly good (and aggressive) about maintaining a persistent hints database for slow MXes. If slow MX processing forms a significant percentage of your spool handling. Exim also has extensive config options for minimising/controlling load distribution on the host system which can be useful for multi-purpose servers or timed load distribution. Postfix used a distributed minumum trust design that's somewhat similar to QMail's. The main differences in its design as compared to QMail is that Postfix compromises less with standard interfaces and file system standards. The author, Wietse Venema, is responsive and helpful, and has a long and largely illustrious history in the security community (author of TCP wrappers etc). The Postfix documentation is thorough but sparse. The config files are easily read, but are not as clear as Exim's. In tests here Postfix's delivery rates were similar to Exim's with a higher initial attack curve (and subsequent higher system loading) but similar values at the inflection points. Postfix's queue handling si similarly intelligent to Exim's with the exception that Postfix does not maintain a slow MX hints DB. I didn't bother evaluating QMail. I've little patience with DJB deliberate abrasiveness, less interest in the violence he delivers to the FHS, and just can't be bothered with his lack of licensing. Not worth the time or bother. Cursory examination of other's stats (such as Amanda's) suggests that its performance and load curves are very similar to Postfix's. -- J C Lawrence ---------(*) Satan, oscillate my metallic sonatas. [EMAIL PROTECTED] He lived as a devil, eh? http://www.kanga.nu/~claw/ Evil is a name of a foeman, as I live. ------------------------------------------------------ Mailman-Users maillist - [EMAIL PROTECTED] http://mail.python.org/mailman/listinfo/mailman-users
