On Thu, 20 Dec 2001 15:12:03 -0800 Kevin Phillips <[EMAIL PROTECTED]> wrote:
> We just did a test run of 50,000 and the server came to a grinding > halt. ooops! What MTA, what hardware, how configured, how tuned? I'm going to get in touch with the technology > manager who handles the Cybergold mailings and find out how they > do it. The very first hack is setting the number of RCPT TOs per message as high as you reasonably can. The exact choice will vary depending on your MTA and MTA's behaviour (sites like AOL filter on over-large RCPT TO lists, and the RFC recommends no more than 100). Past that you start wanting to spread the load about among multiple systems, and to reduce IO loads and contentions on spindles, and so forth. Chuq has commented on his smurf army. I've done similar not-quite-the-same things with domain based routing (different systems dedicated to handling deliveries to specific domain subsets (see BugTraq's list architecture for details/inspiration as well as the LServe docs and so forth) The core problem is that it is inherently an IO constrained problem, and standard MTA's are explicitly NOT built to be friendly to such -- which essentially means you get to write your own MTA (non trivial). -- 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
