>>>>> On Fri, 19 Apr 2002 18:50:48 +0300, >>>>> Nick Ustinov <[EMAIL PROTECTED]> (nu) writes:
nu> Hey everyone, nu> Finally, I have upgraded to 2.1.3. I use skiplist for mailboxes and nu> duplicate db now. Cyrus works together with sendmail and the server is nu> pretty heavily loaded. mailboxes.db file is over 50Mb. The system receives nu> about 150-350 emails per minute, sendmail is set up it deliverymode=q with 1 nu> minute interval. So with Sendmail you've got everything getting queued, then later delivered. nu> At peak times load average was going to 15-20, now it's a bit less in nu> average, however sometimes goes even higher. Anyway -- isn't postfix nu> supposed to be better for such high loads? I've tried to install postfix nu> 1.1.7 (lmtp as mailbox_transport) however when load avg went to 50 I stopped nu> it and returned to sendmail. I know disk speed is essential here -- well, it nu> is fast. But with Postfix it tries to deliver as fast as possible. No wonder you're seeing higher load peaks. Try throttling Postfix to only a certain number of lmtp processes. The request will get queued until load drops off. In the maxproc column for lmtp in master.cf experiment with the numbers until a reasonable balance is struck. On our SMTP gateway we've currently got it set to 4. There's another way to keep all incoming mail queued and then let another process initiate deliveries, but I forget how that's set up. Besides, you may still want to throttle lmtp some. nu> Any comments? Maybe Exim will be better? I guess I will have to put MTA on nu> another machine, otherwise it keeps getting even worse. That's what we've been doing for quite some time now. We do lmtp over private network between the machines. Has worked out well. -- Amos
