A particular reply (not to the list) has told me to try:
" Add to service lmtp: process_min_avail = 20 "
And yes, it works (unless for me), so I want to share it with everybody.
Have not exactly measured performance but this morning have seen over 1400 messages/minute with lmtp under heavy load.
What the "process_min_avail = 20" in service lmtp (or anywhere else) exactly imply? cannot find a reference in wiki2 . I suppose it means than 20 lmtp processes are always open waiting for new connections and so reducing startup latency, isn't it?
Thank you.
On Mon, Oct 17, 2011 at 6:06 AM, Juan Carlos Sanchez <juancarlos.sanchez@upm.es <mailto:juancarlos.sanchez@upm.es>> wrote:
Hello: In the process of migration from version 1.2 to 2.0 have tried to use lmtp instead of lda as delivery agent. But unfortunately througput we get is poor with lmtp compared to lda. In peak times with lda we are able to deliver over 1800 messages/minute (high I/O but not affecting service) but if we switch to lmtp cannot deliver more than 300 mess/minute (very low I/O). This is enough for normal work but low under heavy load (distribution lists messages sometimes stay queued for 4 hours before deliver). We have arround 50000 boxes and use postfix as MTA How to increase lmtp performance? dovecot config for lda and lmtp: protocol lda { auth_socket_path = /usr/local/var/run/dovecot/auth-master info_log_path = /var/log/mail/dovecot_deliver.log log_path = /var/log/mail/dovecot_deliver.log mail_plugins = quota sieve zlib postmaster_address = postmaster@upm.es <mailto:postmaster@upm.es> sendmail_path = /usr/lib/sendmail } protocol lmtp { info_log_path = /var/log/mail/dovecot_deliver.log log_path = /var/log/mail/dovecot_deliver.log mail_plugins = quota sieve zlib postmaster_address = postmaster@upm.es <mailto:postmaster@upm.es> sendmail_path = /usr/lib/sendmail } In postfix I have: mailbox_transport = lmtp:unix:private/dovecot-lmtp Thanks in advance. -- ------------------------------------------------------ Juan Carlos Sanchez Hernandez Universidad Politecnica de Madrid ------------------------------------------------------
--
Juan Carlos Sanchez Hernandez Universidad Politecnica de Madrid
On Lunes 17 Octubre 2011 16:31:01 Juan Carlos Sanchez escribió:
A particular reply (not to the list) has told me to try:
" Add to service lmtp: process_min_avail = 20 "
And yes, it works (unless for me), so I want to share it with everybody.
Have not exactly measured performance but this morning have seen over 1400 messages/minute with lmtp under heavy load.
What the "process_min_avail = 20" in service lmtp (or anywhere else) exactly imply? cannot find a reference in wiki2 . I suppose it means than 20 lmtp processes are always open waiting for new connections and so reducing startup latency, isn't it?
I had a similar problem, and found the same solution. In my setup, only 1 lmtp process was available and a 2nd one never was created, so the process was really busy. With this parameter (I setup it to 10) I have diferent process and all of them are used.
Joseba Torre. Vicegerencia de TICs, área de Explotación
On Mon, 2011-10-17 at 16:42 +0200, Joseba Torre wrote:
Add to service lmtp: process_min_avail = 20
I had a similar problem, and found the same solution. In my setup, only 1 lmtp process was available and a 2nd one never was created, so the process was really busy. With this parameter (I setup it to 10) I have diferent process and all of them are used.
Hmm. I wonder if perhaps it would be better to set:
service lmtp { client_limit = 1 }
I think I'll make that the default..
participants (3)
-
Joseba Torre
-
Juan Carlos Sanchez
-
Timo Sirainen