[Dovecot] M-Box benchmark
A friend of mine passed me this[1] cause I'm recommending him Dovecot.
My personal experience is very different to the one in that benchmark. I will appreciate similar links from those of you who had benchmarked Dovecot against other IMAP servers.
I'm conscious that my personal experience is far to be a objetive comparison. I switched from UW-Imap/Mbox to Dovecot/Maildir a box with 45k local users, 3.5MB/s sustained IMAP/POP traffic in work hours. The main performance gain is from mbox -> maildir switching. Anyway the users (and myself) "perceive" a good performance with the new setup.
Regards, maykel
Maykel Moya wrote:
A friend of mine passed me this[1] cause I'm recommending him Dovecot. [snip] [1] http://www.isode.com/whitepapers/mbox-benchmark.html
My first questions about this test are what version of dovecot was used, did they take into account the fact that dovecot has to build indexes? A 10s ramp-up time seems rather short for that. What kind of auth backend was used for each product and did they even attempt any kind of performance tuning on anything other than their product? This "whitepaper" comes across as extremely biased and very short on important information.
Jonathan
On Mar 17, 2007, at 2:02 PM, Jonathan Stewart wrote:
Maykel Moya wrote:
A friend of mine passed me this[1] cause I'm recommending him
Dovecot. [snip] [1] http://www.isode.com/whitepapers/mbox-benchmark.htmlMy first questions about this test are what version of dovecot was
used, did they take into account the fact that dovecot has to build
indexes? A 10s ramp-up time seems rather short for that. What kind
of auth backend was used for each product and did they even attempt
any kind of performance tuning on anything other than their
product? This "whitepaper" comes across as extremely biased and
very short on important information.
It seems to me that it was a vendor sponsored test, so one would
think they designed the testing criteria to favor their product.
-jav
Javier Henderson wrote:
On Mar 17, 2007, at 2:02 PM, Jonathan Stewart wrote:
Maykel Moya wrote:
A friend of mine passed me this[1] cause I'm recommending him Dovecot. [snip] [1] http://www.isode.com/whitepapers/mbox-benchmark.html
My first questions about this test are what version of dovecot was used, did they take into account the fact that dovecot has to build indexes? A 10s ramp-up time seems rather short for that. What kind of auth backend was used for each product and did they even attempt any kind of performance tuning on anything other than their product? This "whitepaper" comes across as extremely biased and very short on important information.
It seems to me that it was a vendor sponsored test, so one would think they designed the testing criteria to favor their product.
I know and I'm sure they did. I just had to rant a bit because this kind of thing is so common and many people don't even bother to ask questions about it and accept any "study" or "whitepaper" as gospel.
Jonathan
On Sat, 2007-03-17 at 12:59 -0500, Maykel Moya wrote:
This test was done 2-3 years ago with Dovecot 0.99.x version. It doesn't tell anything about Dovecot's current state.
I've thought about doing the test a few times myself, but MStone is a pretty horrible thing to set up.
My current tests show that Dovecot is somewhere around half as fast as Cyrus to twice as fast. It all depends on what the benchmark does.
Quoting Maykel Moya moya@infomed.sld.cu:
comparison. I switched from UW-Imap/Mbox to Dovecot/Maildir a box with 45k local users, 3.5MB/s sustained IMAP/POP traffic in work hours. The main performance gain is from mbox -> maildir switching. Anyway the
Don't be so sure it was from the mbox -> maildir switching. I went from UW-Imap mbox to Dovecot mbox, and the speed up was enormous.
users (and myself) "perceive" a good performance with the new setup.
Yes, all my users perceive it as lightening fast, as do I. How much of a difference depends to some extent on how large your mbox file is...
Regards, maykel
-- Eric Rostetter The Department of Physics The University of Texas at Austin
Go Longhorns!
Hello!
Switching from UW-IMAP/mbox to dovecot/mbox has even a large performance gain (I would say around 10-100, on large folders because of the index files.). Also deleting is much faster.
I haven't benchmarked it but it is definitly much, much faster. (That was the reason I left UW-IMAP).
Ciao, Gerhard
On Sat, 17 Mar 2007, Maykel Moya wrote:
A friend of mine passed me this[1] cause I'm recommending him Dovecot.
My personal experience is very different to the one in that benchmark. I will appreciate similar links from those of you who had benchmarked Dovecot against other IMAP servers.
I'm conscious that my personal experience is far to be a objetive comparison. I switched from UW-Imap/Mbox to Dovecot/Maildir a box with 45k local users, 3.5MB/s sustained IMAP/POP traffic in work hours. The main performance gain is from mbox -> maildir switching. Anyway the users (and myself) "perceive" a good performance with the new setup.
Regards, maykel
On 18.3.2007, at 7.40, Gerhard Wiesinger wrote:
Switching from UW-IMAP/mbox to dovecot/mbox has even a large
performance gain (I would say around 10-100, on large folders
because of the index files.). Also deleting is much faster.I haven't benchmarked it but it is definitly much, much faster. (That was the reason I left UW-IMAP).
I've heard of load average dropping from 100 to 0.02 :)
Timo Sirainen wrote:
On 18.3.2007, at 7.40, Gerhard Wiesinger wrote:
Switching from UW-IMAP/mbox to dovecot/mbox has even a large performance gain (I would say around 10-100, on large folders because of the index files.). Also deleting is much faster.
I haven't benchmarked it but it is definitly much, much faster. (That was the reason I left UW-IMAP).
I've heard of load average dropping from 100 to 0.02 :)
BTW, how is it that you get such great performance using mbox files? I would have thought it would be much slower than it is to delete the first message from say a 2 gig file? How do you do that so fast?
On Sun, 2007-03-18 at 07:22 -0700, Marc Perkel wrote:
I've heard of load average dropping from 100 to 0.02 :)
BTW, how is it that you get such great performance using mbox files? I would have thought it would be much slower than it is to delete the first message from say a 2 gig file? How do you do that so fast?
Expunging the first message from 2GB file is as slow as with UW-IMAP. It's just that usually people don't expunge old mails from large mailboxes. The performance comes simply from not having to read and parse the whole mbox when opening it.
Hello Timo!
UW-IMAP is even slower on EXPUNGE (factor 10 I guess, nearly on every operation).
Dovecot with large files (2GB, e.g. new mails on the end) is really perfect on EXPUNGE (when there are no large holes). It takes <1s to EXPUNGE.
Ciao, Gerhard
On Sun, 18 Mar 2007, Timo Sirainen wrote:
On Sun, 2007-03-18 at 07:22 -0700, Marc Perkel wrote:
I've heard of load average dropping from 100 to 0.02 :)
BTW, how is it that you get such great performance using mbox files? I would have thought it would be much slower than it is to delete the first message from say a 2 gig file? How do you do that so fast?
Expunging the first message from 2GB file is as slow as with UW-IMAP. It's just that usually people don't expunge old mails from large mailboxes. The performance comes simply from not having to read and parse the whole mbox when opening it.
On Sun, 2007-03-18 at 15:55 +0100, Gerhard Wiesinger wrote:
UW-IMAP is even slower on EXPUNGE (factor 10 I guess, nearly on every operation).
Dovecot with large files (2GB, e.g. new mails on the end) is really perfect on EXPUNGE (when there are no large holes). It takes <1s to EXPUNGE.
I guess you mean when expunging new messages? Maybe that's because UW-IMAP again reads and parses the whole mbox before getting to the expunged message, while Dovecot directly begins from that message? Or that's the only thing I can think of. I would have actually expected UW-IMAP to be faster when expunging..
participants (8)
-
Bart Smaalders
-
Eric Rostetter
-
Gerhard Wiesinger
-
Javier Henderson
-
Jonathan Stewart
-
Marc Perkel
-
Maykel Moya
-
Timo Sirainen