[Dovecot] How many concurrent IMAP sessions?
Has anybody ever bothered to measure how many concurrent IMAP sessions dovecot can handle?
I know it depends on hardware, filesystem, mailbox type etc., but what exactly controls the number and what have people on the list measured?
p@rick
- Patrick Ben Koetter p@state-of-mind.de:
Has anybody ever bothered to measure how many concurrent IMAP sessions dovecot can handle?
I know it depends on hardware, filesystem, mailbox type etc., but what exactly controls the number and what have people on the list measured?
We usually have 1200 simultaneous connections during the day, that'S about 8% of our total accounts on the box.
- Ralf Hildebrandt dovecot@dovecot.org:
- Patrick Ben Koetter p@state-of-mind.de:
Has anybody ever bothered to measure how many concurrent IMAP sessions dovecot can handle?
I know it depends on hardware, filesystem, mailbox type etc., but what exactly controls the number and what have people on the list measured?
We usually have 1200 simultaneous connections during the day, that'S about 8% of our total accounts on the box.
On one server, right?
On Wed, 2009-09-30 at 17:08 +0200, Patrick Ben Koetter wrote:
Has anybody ever bothered to measure how many concurrent IMAP sessions dovecot can handle?
I know it depends on hardware, filesystem, mailbox type etc., but what exactly controls the number and what have people on the list measured?
I think disk I/O limit comes first, then memory and finally CPU. I just tested on my desktop:
imaptest clients=1000 - noop=100 delay=100 logout=0
So it basically just issues NOOP commands and does no disk I/O (opens no index files or anything). Looks like each imap process uses about 420 kB of memory (420 MB total according to "free" before/after). That's actually more than I thought. Wonder if I could decrease it..
On Wed, 2009-09-30 at 11:48 -0400, Timo Sirainen wrote:
So it basically just issues NOOP commands and does no disk I/O (opens no index files or anything). Looks like each imap process uses about 420 kB of memory (420 MB total according to "free" before/after). That's actually more than I thought. Wonder if I could decrease it..
Actually imaptest was using about 50 MB and other Dovecot processes maybe 5-10 MB, so it was more like 360 kB per imap process. Slightly better but I think it could still lose 100-200 kB :)
Timo Sirainen wrote:
On Wed, 2009-09-30 at 11:48 -0400, Timo Sirainen wrote:
So it basically just issues NOOP commands and does no disk I/O (opens no index files or anything). Looks like each imap process uses about 420 kB of memory (420 MB total according to "free" before/after). That's actually more than I thought. Wonder if I could decrease it..
Actually imaptest was using about 50 MB and other Dovecot processes maybe 5-10 MB, so it was more like 360 kB per imap process. Slightly better but I think it could still lose 100-200 kB :)
On our live system we've got 1255 concurrent IMAP processes with an average of 732KB private memory each (the rest of the memory should be shared), but they're not (all) idle :)
(using Solaris 8 pmap command: "pgrep imap | xargs -i pmap -x {} | awk '/^total/{tot+=$6;num++}END{print tot, num, tot/num}'")
Best Wishes, Chris
On Wed, 2009-09-30 at 11:51 -0400, Timo Sirainen wrote:
On Wed, 2009-09-30 at 11:48 -0400, Timo Sirainen wrote:
So it basically just issues NOOP commands and does no disk I/O (opens no index files or anything). Looks like each imap process uses about 420 kB of memory (420 MB total according to "free" before/after). That's actually more than I thought. Wonder if I could decrease it..
Actually imaptest was using about 50 MB and other Dovecot processes maybe 5-10 MB, so it was more like 360 kB per imap process. Slightly better but I think it could still lose 100-200 kB :)
Hmm. Now the same calculation shows only about 230kB/process. I don't think I even changed anything..
- Timo Sirainen dovecot@dovecot.org:
On Wed, 2009-09-30 at 11:51 -0400, Timo Sirainen wrote:
On Wed, 2009-09-30 at 11:48 -0400, Timo Sirainen wrote:
So it basically just issues NOOP commands and does no disk I/O (opens no index files or anything). Looks like each imap process uses about 420 kB of memory (420 MB total according to "free" before/after). That's actually more than I thought. Wonder if I could decrease it..
Actually imaptest was using about 50 MB and other Dovecot processes maybe 5-10 MB, so it was more like 360 kB per imap process. Slightly better but I think it could still lose 100-200 kB :)
Hmm. Now the same calculation shows only about 230kB/process. I don't think I even changed anything..
Funny to see you wondering. You always seem to have it all under control. :)
What could cause the variation?
On Qua, 2009-09-30 at 17:08 +0200, Patrick Ben Koetter wrote:
Has anybody ever bothered to measure how many concurrent IMAP sessions dovecot can handle?
I know it depends on hardware, filesystem, mailbox type etc., but what exactly controls the number and what have people on the list measured?
We have about 3600 per server.
-- Jose Celestino SAPO.pt::Systems http://www.sapo.pt --------------------------------------------------------------------- * Progress (n.): The process through which Usenet has evolved from smart people in front of dumb terminals to dumb people in front of smart terminals.
- Jose Celestino japc@co.sapo.pt:
On Qua, 2009-09-30 at 17:08 +0200, Patrick Ben Koetter wrote:
Has anybody ever bothered to measure how many concurrent IMAP sessions dovecot can handle?
I know it depends on hardware, filesystem, mailbox type etc., but what exactly controls the number and what have people on the list measured?
We have about 3600 per server.
Any headroom left or is this the maximum you can get?
p@rick
On Qua, 2009-09-30 at 23:23 +0200, Patrick Ben Koetter wrote:
- Jose Celestino japc@co.sapo.pt:
On Qua, 2009-09-30 at 17:08 +0200, Patrick Ben Koetter wrote:
Has anybody ever bothered to measure how many concurrent IMAP sessions dovecot can handle?
I know it depends on hardware, filesystem, mailbox type etc., but what exactly controls the number and what have people on the list measured?
We have about 3600 per server.
Any headroom left or is this the maximum you can get?
Plenty of headroom still. 8 CPUs, 4GB of ram servers. There are some occasional spikes of load of 2 or 3 because of i/o to the NFS but nothing to worry about. I guess that we could easily take them to 5k concurrent IMAP sessions. But, and this is worth mentioning, this are mostly sessions coming from imapproxy so many of them are being kept by the proxy with no real activity.
-- Jose Celestino SAPO.pt::Systems http://www.sapo.pt --------------------------------------------------------------------- * Progress (n.): The process through which Usenet has evolved from smart people in front of dumb terminals to dumb people in front of smart terminals.
participants (5)
-
Chris Wakelin
-
Jose Celestino
-
Patrick Ben Koetter
-
Ralf Hildebrandt
-
Timo Sirainen