sendmail 8.12.3 dead slow on Solaris 8
Reinier Bezuidenhout
I'm running sendmail on a Solaris 8 system with 2 CPU's 2 Gig RAM etc. I
can't get sendmail to accept more that 10 e-mails per second ? I've tried
to tune a lot of stuff in the sendmail config file, but nothing helped.
I installed smtpd on the machine and I can get 5500 emails per minute.
I've tried using sendmail to only deliver mail too and that also crawls
along only at 10/s.
Is there anything else I can tweak or some compile options ?
Thanks
Reinier
Rich Teer
> I'm running sendmail on a Solaris 8 system with 2 CPU's 2 Gig RAM etc. I
> can't get sendmail to accept more that 10 e-mails per second ? I've tried
> to tune a lot of stuff in the sendmail config file, but nothing helped.
WHat disk sub system are you using?
> I installed smtpd on the machine and I can get 5500 emails per minute.
I've not heard of that one. Presumably, it's a lot more light
weight than sendmail.
--
Rich Teer
President,
Rite Online Inc.
Voice: +1 (250) 979-1638
URL: http://www.rite-online.net
Claus Aßmann
> I'm running sendmail on a Solaris 8 system with 2 CPU's 2 Gig RAM etc. I
> can't get sendmail to accept more that 10 e-mails per second ? I've tried
> to tune a lot of stuff in the sendmail config file, but nothing helped.
What do you mean by "accept"? Where does the mail go after sendmail
accepted it?
> I installed smtpd on the machine and I can get 5500 emails per minute.
Nice. I did 10000 email per minute on an E450 (relaying) with 8.12
in my performance tests. You need a fast disk for that (or turn
off fsync(): does smtpd uses that?)
--
If you feel the urgent wish to send me a courtesy copy of a Usenet
posting, then make sure it's recognizable as such!
The FAQ: http://www.sendmail.org/faq/ Before you ask.
Reinier Bezuidenhout
10000 e-mail per minute ... I would love to get that !!! :)
IF you turn of fsync, would that possibly cause mails to be lost in case of
a system crash ? Is it safe for the file system ?
I'll get the specs on the machine ....
Thanks
Reinier
"Claus Aßmann" <ca+sendmail(-no-copies-please)@mine.informatik.uni-kiel.de>
wrote in message news:aaa639$tte$1...@zardoc.esmtp.org...
Claus Aßmann
[ Don't top-post, see http://www.allmyfaqs.com/faq.pl?How_to_post ]
> IF you turn of fsync, would that possibly cause mails to be lost in case of
> a system crash ? Is it safe for the file system ?
Yes. No.
But some "MTA"s do that anyway to achieve high throughput.
> I'll get the specs on the machine ....
You didn't answer any of my questions:
> "Claus Aßmann" wrote:
> > What do you mean by "accept"? Where does the mail go after sendmail
> > accepted it?
> > Nice. I did 10000 email per minute on an E450 (relaying) with 8.12
> > in my performance tests. You need a fast disk for that (or turn
> > off fsync(): does smtpd uses that?)
It's hard to tell what's going on if you don't provide more data.
Juha Laiho
>I'm running sendmail on a Solaris 8 system with 2 CPU's 2 Gig RAM etc. I
>can't get sendmail to accept more that 10 e-mails per second ? I've tried
>to tune a lot of stuff in the sendmail config file, but nothing helped.
>Is there anything else I can tweak or some compile options ?
Including the .mc file you used to generate your .cf would be nice
(you did tune using the .mc, didn't you?). This would reduce a lot
of guessing about your configuration.
As a specific example, have you set ident timeout (confTO_IDENT) to zero?
What kind of CPU load do you see on the machine when you're getting
that 10 msgs/s? I'd assume the system is running pretty much idle
at that time - and I wouldn't expect a disk bottleneck either.
--
Wolf a.k.a. Juha Laiho Espoo, Finland
(GC 3.0) GIT d- s+: a C++ ULSH++++$ P++@ L+++ E- W+$@ N++ !K w !O !M V
PS(+) PE Y+ PGP(+) t- 5 !X R !tv b+ !DI D G e+ h---- r+++ y++++
"...cancel my subscription to the resurrection!" (Jim Morrison)
Reinier Bezuidenhout
1) accept: I mean receive the e-mail and put it in the queue (sendmail
configured to only queue)
2) After sendmail has received the mail, it delivers it via LMTP to a local
(on the machine) application.
3) smtpd does not use fsync(), but I also turned it off in sendmail with the
SuperSafe=0 option, it still was slow.
4) I've ruled out the disk as being slow since I can get smtpd to receive
(write into a queue) 5500 e-mails per minute.
I know that sendmail on solaris 5.8 uses fcntl to lock files rather than
flock, might that be the reason ?
Thanks
Reinier
"Claus Aßmann" <ca+sendmail(-no-copies-please)@mine.informatik.uni-kiel.de>
wrote in message news:aabuhq$js1$1...@zardoc.esmtp.org...
Reinier Bezuidenhout
Below is the .mc file I used to create the .cf file ...
I turned ident off in the .cf and I also tried SuperSafe=0 to disable fsync.
the load on the system is about 60% iowait and 30% idle ... the iowait,
would that indicate a disk bottlenect or does it also list iowait if it
waits on a socket ?
Thanks
Reinier
divert(0)dnl
VERSIONID(`$Id: generic-sunos4.1.mc,v 8.11 1999/02/07 07:26:03 gshapiro Exp
$')
OSTYPE(solaris2)dnl
FEATURE(local_lmtp)dnl
FEATURE(promiscuous_relay)dnl
FEATURE(accept_unresolvable_domains)dnl
define(`confDELIVERY_MODE', `queue')dnl
define(`confSERVICE_SWITCH_FILE',`/etc/mail/nsswitch.conf')dnl
define(`confMAX_DAEMON_CHILDREN',`30')dnl
define(`confBIND_OPTS', `-DNSRCH -DEFNAMES')dnl
FEATURE(nocanonify)dnl
DOMAIN(generic)dnl
MAILER(local)dnl
MAILER(smtp)dnl
"Juha Laiho" <Juha....@iki.fi> wrote in message
news:aabv58$f0a$1...@ichaos.ichaos-int...
Rich Teer
What part of "please don't top post" don't you understand?!
> divert(0)dnl
> VERSIONID(`$Id: generic-sunos4.1.mc,v 8.11 1999/02/07 07:26:03 gshapiro Exp
> $')
> OSTYPE(solaris2)dnl
That should be solaris8.
> FEATURE(local_lmtp)dnl
> FEATURE(promiscuous_relay)dnl
Please tell us your IP address, so that we can add you
to our black lists...
Claus Aßmann
[once again: don't top-post, see
http://www.allmyfaqs.com/faq.pl?How_to_post ]
> 1) accept: I mean receive the e-mail and put it in the queue (sendmail
> configured to only queue)
> 2) After sendmail has received the mail, it delivers it via LMTP to a local
> (on the machine) application.
So you perform queue runs?
> 3) smtpd does not use fsync(), but I also turned it off in sendmail with the
> SuperSafe=0 option, it still was slow.
> 4) I've ruled out the disk as being slow since I can get smtpd to receive
> (write into a queue) 5500 e-mails per minute.
Yes, but without fsync(). What disk is this, which FS? The
conventional Sun-FS with their SCSI disks is very slow for an MTA
application.
If you want similar test conditions, then don't run the queue:
sendmail -bd -odd -O SuperSafe=Off
Run your test and compare.
If that's sufficiently fast, try:
sendmail -bd -odd
sendmail -bd -odq
and finally add local delivery back. I assume the last
step is the "killer".
As long as you are comparing different things the results
will be useless.
Reinier Bezuidenhout
"Rich Teer" <ri...@rite-group.com> wrote in message
news:Pine.GSO.4.44.020426...@grover.rite-group.com...
> On Fri, 26 Apr 2002, Reinier Bezuidenhout wrote:
>
> What part of "please don't top post" don't you understand?!
This part ...
>
> > divert(0)dnl
> > VERSIONID(`$Id: generic-sunos4.1.mc,v 8.11 1999/02/07 07:26:03 gshapiro
Exp
> > $')
> > OSTYPE(solaris2)dnl
>
> That should be solaris8.
>
> > FEATURE(local_lmtp)dnl
> > FEATURE(promiscuous_relay)dnl
>
> Please tell us your IP address, so that we can add you
> to our black lists...
>
These features were added to see if the rule processing had any impact on
the speed ...
Thanks
Reinier
Reinier Bezuidenhout
"Claus Aßmann" <ca+sendmail(-no-copies-please)@mine.informatik.uni-kiel.de>
wrote in message news:aac81q$56d$1...@zardoc.esmtp.org...
> Reinier Bezuidenhout wrote:
>
> [once again: don't top-post, see
> http://www.allmyfaqs.com/faq.pl?How_to_post ]
>
> So you perform queue runs?
Yes .. but when I did the performance tests I disabled the queue runs in the
.cf file by setting delivery mode to defer or queue
>
> Yes, but without fsync(). What disk is this, which FS? The
> conventional Sun-FS with their SCSI disks is very slow for an MTA
> application.
>
I'm sure it is Sun-FS with SCSI disks ... what other options do I have ?
> If you want similar test conditions, then don't run the queue:
>
> sendmail -bd -odd -O SuperSafe=Off
File size = 5728
time,messages,data(K),errors,connections
15:38,582,3496,0,119,0 ---> 582 e-mails / minute
>
> Run your test and compare.
> If that's sufficiently fast, try:
> sendmail -bd -odd
File size = 5728
time,messages,data(K),errors,connections
15:40,599,3604,0,120,0 ----> 599 e-mails
> sendmail -bd -odq
> and finally add local delivery back. I assume the last
> step is the "killer".
>
> As long as you are comparing different things the results
> will be useless.
Not even with all the options I can get more thant about 10 e-mails /
second. When doing a "netstat -an" there is no connections to DNS since
I've disabled DNS lookups.
Thanks for your time !!
Reinier
Claus Aßmann
> "Claus Aßmann"
> > Reinier Bezuidenhout wrote:
> > Yes, but without fsync(). What disk is this, which FS? The
> > conventional Sun-FS with their SCSI disks is very slow for an MTA
> > application.
> I'm sure it is Sun-FS with SCSI disks ... what other options do I have ?
My test machine had a Baydel (?) RAID system...
> > If you want similar test conditions, then don't run the queue:
> > sendmail -bd -odd -O SuperSafe=Off
> File size = 5728
> time,messages,data(K),errors,connections
> 15:38,582,3496,0,119,0 ---> 582 e-mails / minute
Ok, I tried this on another Sun... their FS is extremely slow.
Removing 1200 files takes 26 seconds... That's ridiculous.
It seems meta-data operations are very expensive.
Seems you need to get a better FS... ask Sun.
Don't they have something else to offer (LVM? Veritas?)
Rich Teer
> Ok, I tried this on another Sun... their FS is extremely slow.
> Removing 1200 files takes 26 seconds... That's ridiculous.
> It seems meta-data operations are very expensive.
UFS logging might help here.
Claus Aßmann
> On 26 Apr 2002, Claus Aßmann wrote:
> > Ok, I tried this on another Sun... their FS is extremely slow.
> > Removing 1200 files takes 26 seconds... That's ridiculous.
> > It seems meta-data operations are very expensive.
> UFS logging might help here.
Does Sun support a logging (journaling) FS?
If so, which man pages are relevant?
I can't find it on my SunOS 5.8 box :-(
Tony Curtis
>> Claus Aßmann <ca+sendmail(-no-copies-please)@mine.informatik.uni-kiel.de> said:
> Rich Teer wrote:
>> On 26 Apr 2002, Claus Aßmann wrote:
>> UFS logging might help here.
> Does Sun support a logging (journaling) FS?
Yes, ufs.
> If so, which man pages are relevant? I can't find it on
> my SunOS 5.8 box :-(
mount_ufs(1M)
hth
t
--
Wait up guys! I've gotta empty my grass bag.
Rich Teer
> Does Sun support a logging (journaling) FS?
Yep; UFS has supported logging since Solaris 7.
It was available with prior releases if you used
DiskSuite.
> If so, which man pages are relevant?
> I can't find it on my SunOS 5.8 box :-(
mount_ufs
HTH,
Reinier Bezuidenhout
"Claus Aßmann" <ca+sendmail(-no-copies-please)@mine.informatik.uni-kiel.de>
wrote in message news:aaflt1$amt$2...@zardoc.esmtp.org...
> Rich Teer wrote:
> > On 26 Apr 2002, Claus Aßmann wrote:
>
> > > Ok, I tried this on another Sun... their FS is extremely slow.
> > > Removing 1200 files takes 26 seconds... That's ridiculous.
> > > It seems meta-data operations are very expensive.
>
> > UFS logging might help here.
I enabled logging and noatime on the UFS and it tripled the throughput ...
this is a start .. 28 / sec thanks !!!!
:) (all smiles here)