Re: load average question

2002-11-25 Thread Jeremy C. Reed
On Mon, 25 Nov 2002, Sonny Kupka wrote:

> pop-3   stream  tcp nowait.200  root/usr/sbin/tcpd 
> /usr/sbin/in.qpopper -R -B -f /etc/qpopper.conf
> 
> Key being: nowait.200

xinetd uses different configurations than inetd.

Some parts of my configs:

# - added cps for 50 seconds and wait 15 seconds
# - added 200 processes running at a time
# - added --timeout 1200 (due to bad clients that don't close session
# and keep ESTABLISHED state alive)
instances   = 200
cps = 50 15
server  = /usr/local/sbin/vm-pop3d
server_args = -i --timeout 1200

cps is the incoming connection limiter: connections per second to handle
(and number of seconds to wait before re-enabling after disabled).

instances if for simultaneous active servers.

xinetd also can be customized for limits per source IP and load average
and more.

(But inetd works well for me.)

  Jeremy C. Reed
..
 ISP-FAQ.com -- find answers to your questions
 http://www.isp-faq.com/


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]




Re: load average question

2002-11-25 Thread Olivier Macchioni
++ 25/11/02 08:51 -0800 - Jeremy C. Reed:
>On Sat, 23 Nov 2002, Scott wrote:
>
>> > No performance issues using vm-pop3d, exim (MTA), apache and
>> > OpenWebMail with around 10,000 email accounts on similar hardware.
>> 
>> 10,000 email accounts on a similar machine?  Man, I must be doing 
>> something wrong!  
[...]
>> I tried that and the qpopper stopped responding after about 30 minutes and 
>> the Outlook users would get the xinetd error.  

Please don't use inetd / xinetd if you're looking for performance. Both
spawn a new process each time there is an incoming request, which is a
*huge* overhead.

You definitely have to run qpopper (or any other POP daemon) as a
standalone server. 

Other tips include :
- recompiling qpopper with other options (no debugging, optimization)
- reading http://www.eudora.com/qpopper/faq.html#performance

and keep in mind that the number of mailboxes is not a significant
number to compare performance. I'd rather look at the number of
connections / minute.


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]




Re: load average question

2002-11-25 Thread Jeremy C. Reed
On Sat, 23 Nov 2002, Scott wrote:

> > No performance issues using vm-pop3d, exim (MTA), apache and
> > OpenWebMail with around 10,000 email accounts on similar hardware.
> 
> 10,000 email accounts on a similar machine?  Man, I must be doing 
> something wrong!  I assume you are running Debian?  Sounds more and more 

In fact, that similar machine is Red Hat Linux 7.2 on i386 hardware.

(But have also done similar with Debian, BSD/OS and others.)

> to me like a Red Hat and sendmail problem in my case if I can't handle 
> 2,300 mail accounts.

Red Hat Linux and Debian Linux should have near same performance. (A
difference could be the filesystem used and how it is used.)

> > In the past, when using qpopper with 10-15,000 accounts, I improved
> > performance by using qpopper "server" mode.
> 
> I tried that and the qpopper stopped responding after about 30 minutes and 
> the Outlook users would get the xinetd error.  

What xinetd error?

(You can configure xinetd for many connections.)

> Thank you to everyone who responded, it is obvious that I need to get a 
> Debian box up and see how it performs under this load, I am sure it will 
> do much better.

I don't see how the performance could be much different when using the
same software.

(But many find that Debian is usually easier to administer.)

  Jeremy C. Reed
..
 ISP-FAQ.com -- find answers to your questions
 http://www.isp-faq.com/


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]




Re: load average question

2002-11-25 Thread Nick Mitchell
I am running Debian 3.0 on one mail server with 3,000+ mailboxes using
Postfix/Procmail/IMP/MySQL/qpopper/Apache. And I have very low cpu/mem
usages. And it is a single proc PIII-500 with 256MB of ram. And it works
great. 

I agree with the other posts that qpopper is a dog, but I don't such a
performance degrade that it is worth changing this far in the game.

Nick



On Fri, 22 Nov 2002 17:51:19 -0500
"Scott St. John" <[EMAIL PROTECTED]> wrote:

> Hi Gang:
> 
> A few weeks ago we talked about me moving a server from BSDi to Debian.  As 
> luck
> would have it that BSDi server gave out last Monday and I had to move fast 
> to replace
> it.  Knowing I can do a RH install in about 30 minutes I went the route of 
> familiar
> territory and installed 7.2 with Sendmail/QPopper/Apache/OpenWebMail.  I am 
> paying
> for that now with a huge performance problem.  I am seeing Load Averages 
> spiking
> above 6 during the day.  Hardware is a Dual P3-600 with a gig of ram on a IBM
> Netfinity Raid 5 controller.
> 
> The owner of the company wants to go back to BSD, but I want to pursue Debian.
> So the question is:  is anyone running a similar set up with either Sendmail or
> Posrtfix servicing 2,000+ email accounts with any performance issues?
> 
> Thank you for your time.
> 
> -Scott

~~~
Nick Mitchell
Network Engineer
Delaware.Net, Inc.
Mail: [EMAIL PROTECTED]
Voice:302.736.5515 x17
Web: http://www.delaware.net
ICQ: 31676269
Artificial Intelligence: The Final Frontier
~~~


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]




Re: load average question

2002-11-25 Thread Sonny Kupka


 Qpopper also has a habit of just stopping, I have have to
reload xinetd several times a day.


I run qpopper here with users keeping mail in /var/mail/username and have 
no problems what so ever with the setup..

You have your server set to allow alot of "pops" at the same time?

I have this:

pop-3   stream  tcp nowait.200  root/usr/sbin/tcpd 
/usr/sbin/in.qpopper -R -B -f /etc/qpopper.conf

Key being: nowait.200

Oh you said this was a redsplat box right now..

I never ran that distro.. went from slackwhore to Debian so I hope it's 
about the same :)

---
Sonny





Had a Debian box ready to go, but when the machine crashed I had to get
something up right then and now so I went with RH because I could do it
in under 30 minutes, plus I had the Sendmail configs.  I ended up using
Suse as well for Cistron Radius.  I only have 3 of these boxes - Mandrake
for the web server, RH for the mail and Suse sits there and does Radius.
I have been planning on moving people off the Mandrake box so I might
move them to the Suse box since it only does Radius, turn the Mandrake
box into Debian and be happy :)

-Scott



---
Outgoing mail is certified Virus Free.
Checked by AVG anti-virus system (http://www.grisoft.com).
Version: 6.0.419 / Virus Database: 235 - Release Date: 11/13/2002



--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]




Re: load average question

2002-11-24 Thread Cameron Moore
* [EMAIL PROTECTED] (Jeremy C. Reed) [2002.11.23 17:19]:
> On Fri, 22 Nov 2002, Scott St. John wrote:
> > So the question is:  is anyone running a similar set up with either
> > Sendmail or Posrtfix servicing 2,000+ email accounts with any
> > performance issues?
> 
> No performance issues using vm-pop3d, exim (MTA), apache and
> OpenWebMail with around 10,000 email accounts on similar hardware.
> 
> In the past, when using qpopper with 10-15,000 accounts, I improved
> performance by using qpopper "server" mode.

The number of email accounts is a false indicator.  How many messages do
you receive each day?  What are your mesgs/sec statistics under normal
load?  How active is the webmail application?

Could you be more specific about the disk/raid setup?

I'm asking all these questions because I'm going to be replacing a mail
server soon with 6000+ accounts receiving about 80K mesgs per day.  I'm
curious to hear about other setups and they loads they can sustain.
Thanks
-- 
Cameron Moore
[ Why is the word dictionary in the dictionary? ]


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]




Re: load average question

2002-11-24 Thread Craig Sanders
On Sat, Nov 23, 2002 at 11:01:28PM -0500, Scott wrote:
> On Sun, 24 Nov 2002, Craig Sanders wrote:
> >  - install radius on the RH or Mandrake machine
> >  - configure your NAS boxes to use it
> >  - blow away the suse box and install debian 
> >  - (optionally) move radius back to it, configure NAS boxes to use it again.
> >  - install apache and start moving vhost clients over
> 
> Uh, considering I am running on little sleep the last week, yes that
> would make a lot more sense.  We encountered several problems getting
> radius to start on the Red Hat box and I don't trust the Mandrake box
> to handle it, I consider radius mission critical and I don't trust
> Mandrake in that role any more.  

a quick solution here is to put a working drive in the dead BSDi box(*)
and install debian and cistron-radius on it.  IDE or SCSI would be fine,
it's only temporary to give you another machine to juggle services
around. 

btw, in my experience, cistron radius on debian has been rock-solid
stable and reliable.  i'll probably end up switching to freeradius
(still on debian) in the future, but not until either a) i really need
some of the new features or b) there's a non-beta release of freeradius.  

(*) i think you said it was the drive that failed here.

craig

-- 
craig sanders <[EMAIL PROTECTED]>

Fabricati Diem, PVNC.
 -- motto of the Ankh-Morpork City Watch


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]




Re: load average question

2002-11-23 Thread Scott
On Sun, 24 Nov 2002, Craig Sanders wrote:
>  - install radius on the RH or Mandrake machine
>  - configure your NAS boxes to use it
>  - blow away the suse box and install debian 
>  - (optionally) move radius back to it, configure NAS boxes to use it again.
>  - install apache and start moving vhost clients over

Uh, considering I am running on little sleep the last week, yes that would 
make a lot more sense.  We encountered several problems getting radius to 
start on the Red Hat box and I don't trust the Mandrake box to handle it, 
I consider radius mission critical and I don't trust Mandrake in that role 
any more.  Of course we were trying with Livingston radius on Red Hat, but 
ended up using Cistron on the Suse box.  I might give Cistron a go on Red 
Hat and see if I can move radius to that box and then install Debian on 
the current Suse box.



-Scott



-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]




Re: load average question

2002-11-23 Thread Scott
On Sat, 23 Nov 2002, Jeremy C. Reed wrote:
> Use top(1) to to view processes.

Have been doing that constantly, my eyes hurt :)  Sendmail the top dog 
with the occasional hit from Openwebmail which shows as perl.  Popper will 
do a quick peak and then usually drop off.

> No performance issues using vm-pop3d, exim (MTA), apache and
> OpenWebMail with around 10,000 email accounts on similar hardware.

10,000 email accounts on a similar machine?  Man, I must be doing 
something wrong!  I assume you are running Debian?  Sounds more and more 
to me like a Red Hat and sendmail problem in my case if I can't handle 
2,300 mail accounts.

> In the past, when using qpopper with 10-15,000 accounts, I improved
> performance by using qpopper "server" mode.

I tried that and the qpopper stopped responding after about 30 minutes and 
the Outlook users would get the xinetd error.  

Thank you to everyone who responded, it is obvious that I need to get a 
Debian box up and see how it performs under this load, I am sure it will 
do much better.

-Scott



-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]




Re: load average question

2002-11-23 Thread Jeremy C. Reed
On Fri, 22 Nov 2002, Scott St. John wrote:

> territory and installed 7.2 with Sendmail/QPopper/Apache/OpenWebMail.  I am 
> paying
> for that now with a huge performance problem.  I am seeing Load Averages 
> spiking
> above 6 during the day.  Hardware is a Dual P3-600 with a gig of ram on a IBM
> Netfinity Raid 5 controller.

Use top(1) to to view processes.

With Red Hat, using the sadc (system activity data collector) with sa and
sar can easily tell you wwhat processes/programs are your resource hogs.

(Your BSD/OS had similar with lastcomm and sa.)

> The owner of the company wants to go back to BSD, but I want to pursue
> Debian.

Debian is great. NetBSD is also great and very similar to BSD/OS (as
another alternative).

> So the question is:  is anyone running a similar set up with either
> Sendmail or Posrtfix servicing 2,000+ email accounts with any
> performance issues?

No performance issues using vm-pop3d, exim (MTA), apache and
OpenWebMail with around 10,000 email accounts on similar hardware.

In the past, when using qpopper with 10-15,000 accounts, I improved
performance by using qpopper "server" mode.

Good luck,

  Jeremy C. Reed
echo 'G014AE824B0-07CC?/JJFFFI?D64CB>D=3C427=>;>6HI2>


Re: load average question

2002-11-23 Thread Nathan E Norman
On Sat, Nov 23, 2002 at 09:20:25AM +0100, Torsten Krueger wrote:
> Hi,
> 
> On Sat, 23 Nov 2002, Russell Coker wrote:
> 
> > On Fri, 22 Nov 2002 23:51, Scott St. John wrote:
> > > A few weeks ago we talked about me moving a server from BSDi to Debian.  As
> > > luck
> > > would have it that BSDi server gave out last Monday and I had to move fast
> > > to replace
> > > it.  Knowing I can do a RH install in about 30 minutes I went the route of
> > > familiar
> > > territory and installed 7.2 with Sendmail/QPopper/Apache/OpenWebMail.  I am
> > > paying
> > > for that now with a huge performance problem.  I am seeing Load Averages
> > > spiking
> > > above 6 during the day.  Hardware is a Dual P3-600 with a gig of ram on a
> > > IBM Netfinity Raid 5 controller.
> 
> Hmm - watch out for the load qpopper produces. Since the sendmail/qpopper
> combination uses mbox a users mailbox is copied every time the mbox is
> accessed. If you have large mailboxes (e.g. no quota an users leave mail
> on server) this can produce a significant load. 



In a former life (4+ years ago) we had to move from qpopper to
cucipop for exactly this reason.  Quotas help too.

These days I think the servers in question are running qmail-pop3d.

-- 
Nathan Norman - Incanus Networking mailto:[EMAIL PROTECTED]
  Avoid gunfire in the bathroom tonight.


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]




Re: load average question

2002-11-23 Thread Craig Sanders
On Sat, Nov 23, 2002 at 07:56:15AM -0500, Scott St. John wrote:
> At 01:43 PM 11/23/2002 +0100, Torsten Krueger wrote:
> >You can convert the mboxes with mbox2maildir. Changing from Sendmail
> >to Postfix shouldn't bei a hassle and if properly prepared shouldn't
> >produce a long downtime. Depending on the amount and size of your
> >mboxes the conversion takes some time, but your machine should be
> >fast enough ;-) to do this in a reasonable time.

personally, i'd do the mbox->maildir conversion as one of the last steps
after postfix has been tested and shown to be working well for (at
least!) several days, purely so that the whole migration is a steady
sequence of small incremental improvements, each of which can be
performed and tested in turn.

trying to change too much in one go can lead to a real mess.

the key to successful migration is to plan out exactly what you're going
to do, in what order, and test each step as it is completed.  if nothing
else, it is very useful to figure out the optimum path (i.e. least
hassle, least downtime, least after-the-fact realising "damn, i should
have done this other step before this one", etc) from where you are now
to where you want to be.

> Since I had a Suse box doing nothing but Radius I am planning on
> moving the virtual hosting clients on the Mandrake box to Suse, then

wouldn't it make more sense to:

 - install radius on the RH or Mandrake machine
 - configure your NAS boxes to use it
 - blow away the suse box and install debian 
 - (optionally) move radius back to it, configure NAS boxes to use it again.
 - install apache and start moving vhost clients over

that way you end up with a debian web server rather than suse.  much
better.

> install Debian on the former Mandrake box and make that the mail
> server.  

yep.

also, you might want to eventually move radius to this box too.  if
you're not using LDAP or other shared account db, it can be very useful
to have with the radius server on the same machine as the mail server.
/etc/passwd already holds the passwords for login accounts, it's easy to
configure radius to use that.  

craig

-- 
craig sanders <[EMAIL PROTECTED]>

Fabricati Diem, PVNC.
 -- motto of the Ankh-Morpork City Watch


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]




Re: load average question

2002-11-23 Thread Marc Haber
On Sat, 23 Nov 2002 22:57:45 +1100, Craig Sanders <[EMAIL PROTECTED]>
wrote:
>Maildir does have some other advantages (e.g. it's NFS safe, and there
>are no locking hassles*) so, all else being equal, it's a good choice to
>make...but mbox isn't a bad choice either.

There is no decent IMAP server available that can use mbox format.
This was our reason to use maildir on all new systems.

courier-IMAP does not rule, but it's clearly the least evil.

Greetings
Marc

-- 
-- !! No courtesy copies, please !! -
Marc Haber  |   " Questions are the | Mailadresse im Header
Karlsruhe, Germany  | Beginning of Wisdom " | Fon: *49 721 966 32 15
Nordisch by Nature  | Lt. Worf, TNG "Rightful Heir" | Fax: *49 721 966 31 29


--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]




Re: load average question

2002-11-23 Thread Adrian 'Dagurashibanipal' von Bidder
On Fri, 2002-11-22 at 23:51, Scott St. John wrote:


> territory and installed 7.2 with Sendmail/QPopper/Apache/OpenWebMail.  I am 

Of course, people on the postfix mailing list are biased  but I
think switching away from sendmail would take off some of the load
(assuming that sendmail actually *is* the performance hog.)

cheers
-- vbi

-- 
this email is protected by a digital signature:  http://fortytwo.ch/gpg

NOTE: keyserver bugs! get my key here: https://fortytwo.ch/gpg/92082481





signature.asc
Description: This is a digitally signed message part


Re: load average question

2002-11-23 Thread Scott St. John
At 01:43 PM 11/23/2002 +0100, Torsten Krueger wrote:

You can convert the mboxes with mbox2maildir. Changing from Sendmail to
Postfix shouldn't bei a hassle and if properly prepared shouldn't produce
a long downtime. Depending on the amount and size of your mboxes the
conversion takes some time, but your machine should be fast enough ;-) to
do this in a reasonable time.


Since I had a Suse box doing nothing but Radius I am planning on moving the
virtual hosting clients on the Mandrake box to Suse, then install Debian on the
former Mandrake box and make that the mail server.  Sounds confusing and
it is :)  Then I will try to get Radius running on the Debian box which will
free up two more boxes.  Guess what I will be doing Thanksgiving week :)


Hmm - maintaining such a distribution zoo sounds like a headache. Switch
all the boxes to debian, install a local apt-proxy and enjoy applying
security patches to all your machines in minutes ;-)


It is, but I had to go with familiar to stop the downtime.  I knew I could 
get Radius
going on Suse, I knew I could build a quick RH mail server and over the summer
I *thought* Mandrake was a decent web server.

-Scott

---
Outgoing mail is certified Virus Free.
Checked by AVG anti-virus system (http://www.grisoft.com).
Version: 6.0.419 / Virus Database: 235 - Release Date: 11/13/2002



Re: load average question

2002-11-23 Thread Torsten Krueger
Hi Scott,

On Sat, 23 Nov 2002, Scott St. John wrote:

> Is it possible to move current mail to the maildir format?  I inconvenienced
> the clients once with the crash and now that things are not running as smooth
> I don't want to run the risk of long amounts of downtime.  I really want to go
> Postfix because I think it is more efficient, more secure, more reliable 
> and less
> of a resource pig.  Qpopper also has a habit of just stopping, I have have to
> reload xinetd several times a day.

You can convert the mboxes with mbox2maildir. Changing from Sendmail to
Postfix shouldn't bei a hassle and if properly prepared shouldn't produce
a long downtime. Depending on the amount and size of your mboxes the
conversion takes some time, but your machine should be fast enough ;-) to
do this in a reasonable time.

> 
> 
> >Regarding a switch to debian. Three years ago we started switching our
> >servers from Suse to Debian. I've never regretted that switch since
> >maintaining Debian is so much easier than Suse due to apt.
> 
> Had a Debian box ready to go, but when the machine crashed I had to get
> something up right then and now so I went with RH because I could do it
> in under 30 minutes, plus I had the Sendmail configs.  I ended up using
> Suse as well for Cistron Radius.  I only have 3 of these boxes - Mandrake
> for the web server, RH for the mail and Suse sits there and does Radius.
> I have been planning on moving people off the Mandrake box so I might
> move them to the Suse box since it only does Radius, turn the Mandrake
> box into Debian and be happy :)
> 
Hmm - maintaining such a distribution zoo sounds like a headache. Switch
all the boxes to debian, install a local apt-proxy and enjoy applying
security patches to all your machines in minutes ;-)

Regards
Torsten Krueger


-- 
Media Online Internet Services & Marketing GmbH
Torsten Krueger   [EMAIL PROTECTED]
fon: 49-231-5575100fax: 49-231-55751098
Ruhrallee 39   D-44137 Dortmund


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]




Re: load average question

2002-11-23 Thread Scott St. John
At 09:20 AM 11/23/2002 +0100, Torsten Krueger wrote:

Hmm - watch out for the load qpopper produces. Since the sendmail/qpopper
combination uses mbox a users mailbox is copied every time the mbox is
accessed. If you have large mailboxes (e.g. no quota an users leave mail
on server) this can produce a significant load.


We require that users download their mail and not store it on the server.  We
do have a lot of business accounts where they transfer large files, but for the
most part most of the mail boxes are small - under the 1 meg size.

>I would see no problems with that. If you can make sure that the popper

produces the load I'd suggest moving to Maildir for storage of the users
mailboxes and perhaps Postfix with Courier Pop3d.


Is it possible to move current mail to the maildir format?  I inconvenienced
the clients once with the crash and now that things are not running as smooth
I don't want to run the risk of long amounts of downtime.  I really want to go
Postfix because I think it is more efficient, more secure, more reliable 
and less
of a resource pig.  Qpopper also has a habit of just stopping, I have have to
reload xinetd several times a day.


Regarding a switch to debian. Three years ago we started switching our
servers from Suse to Debian. I've never regretted that switch since
maintaining Debian is so much easier than Suse due to apt.


Had a Debian box ready to go, but when the machine crashed I had to get
something up right then and now so I went with RH because I could do it
in under 30 minutes, plus I had the Sendmail configs.  I ended up using
Suse as well for Cistron Radius.  I only have 3 of these boxes - Mandrake
for the web server, RH for the mail and Suse sits there and does Radius.
I have been planning on moving people off the Mandrake box so I might
move them to the Suse box since it only does Radius, turn the Mandrake
box into Debian and be happy :)

-Scott


---
Outgoing mail is certified Virus Free.
Checked by AVG anti-virus system (http://www.grisoft.com).
Version: 6.0.419 / Virus Database: 235 - Release Date: 11/13/2002



Re: load average question

2002-11-23 Thread Scott St. John
At 08:40 AM 11/23/2002 +0100, Russell Coker wrote:

Apart from webmail that should be a trivial load.  Webmail systems seem to
take up lots of resources in my experience, is it an option to have a
separate machine for webmail?


I was thinking the same thing, if I can get Openwebmail to load on another 
machine
and have it check the mailboxes on the mail server it might help.

-Scott

---
Outgoing mail is certified Virus Free.
Checked by AVG anti-virus system (http://www.grisoft.com).
Version: 6.0.419 / Virus Database: 235 - Release Date: 11/13/2002



Re: load average question

2002-11-23 Thread Craig Sanders
On Sat, Nov 23, 2002 at 10:26:14AM +0100, Russell Coker wrote:
> On Sat, 23 Nov 2002 09:20, Torsten Krueger wrote:
> > Hmm - watch out for the load qpopper produces. Since the
> > sendmail/qpopper combination uses mbox a users mailbox is copied
> > every time the mbox is accessed. If you have large mailboxes (e.g.
> > no quota an users leave mail on server) this can produce a
> > significant load.
> 
> You are correct, any pop server which uses mbox stores will cause very
> high system load, and qpopper is worse than some other pop servers.
> 
> You and Craig are right about the benefits of Maildir.

that's partly true.  it depends entirely on usage patterns, and not all
mbox pop daemons are as bad as qpopper.  some are quite good.

for some usage patterns (i.e. large mailboxes left on the server on a
decent filesystem like XFS or reiserfs), Maildir is a clear winner.

for other usage patterns, the advantage isn't so clear - and in some
cases, it loses out to mbox.  e.g. it's faster to open one file and read
it in one go than it is to have lots of little files, opening, reading,
and then closing them one-by-one.


Maildir does have some other advantages (e.g. it's NFS safe, and there
are no locking hassles*) so, all else being equal, it's a good choice to
make...but mbox isn't a bad choice either.

mbox is also, IMO, a lot more convenient than Maildir when you use mutt
(or elm, pine, etc) - it's easier to manipulate groups of mbox files
than Maildir directories with squillions of little files in them.


* mbox locking isn't a problem on debian systems anyway.  all 
programs that read or write mbox files conform to policy so locking
actually works reliably.  still doesn't make mbox NFS-safe, though.



in short, if you can't choose between mbox and Maildir, then toss a
coin.  whichever way it comes up will work fine :)


craig

-- 
craig sanders <[EMAIL PROTECTED]>

Fabricati Diem, PVNC.
 -- motto of the Ankh-Morpork City Watch


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]




Re: load average question

2002-11-23 Thread Russell Coker
On Sat, 23 Nov 2002 09:20, Torsten Krueger wrote:
> Hmm - watch out for the load qpopper produces. Since the sendmail/qpopper
> combination uses mbox a users mailbox is copied every time the mbox is
> accessed. If you have large mailboxes (e.g. no quota an users leave mail
> on server) this can produce a significant load.

You are correct, any pop server which uses mbox stores will cause very high 
system load, and qpopper is worse than some other pop servers.

I was just looking at the concept rather than the details when I wrote my 
previous quick email.

You and Craig are right about the benefits of Maildir.

> I would see no problems with that. If you can make sure that the popper
> produces the load I'd suggest moving to Maildir for storage of the users
> mailboxes and perhaps Postfix with Courier Pop3d.

I agree totally.

> > Apart from webmail that should be a trivial load.  Webmail systems seem
> > to take up lots of resources in my experience, is it an option to have a
> > separate machine for webmail?
> >
> > http://www.coker.com.au/postal/Postal SMTP/POP benchmark
>
> Try this one and look what type of the system is the reason for the
> problems.

;)

-- 
http://www.coker.com.au/selinux/   My NSA Security Enhanced Linux packages
http://www.coker.com.au/bonnie++/  Bonnie++ hard drive benchmark
http://www.coker.com.au/postal/Postal SMTP/POP benchmark
http://www.coker.com.au/~russell/  My home page


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]




Re: load average question

2002-11-23 Thread Craig Sanders
On Sat, Nov 23, 2002 at 08:40:20AM +0100, Russell Coker wrote:
> On Fri, 22 Nov 2002 23:51, Scott St. John wrote:
> > A few weeks ago we talked about me moving a server from BSDi to
> > Debian.  As luck would have it that BSDi server gave out last Monday
> > and I had to move fast to replace it.  Knowing I can do a RH install
> > in about 30 minutes I went the route of familiar territory and
> > installed 7.2 with Sendmail/QPopper/Apache/OpenWebMail.  I am paying
> > for that now with a huge performance problem.  I am seeing Load
> > Averages spiking above 6 during the day.  Hardware is a Dual P3-600
> > with a gig of ram on a IBM Netfinity Raid 5 controller.
> >
> > The owner of the company wants to go back to BSD, but I want to
> > pursue Debian. So the question is:  is anyone running a similar set
> > up with either Sendmail or Posrtfix servicing 2,000+ email accounts
> > with any performance issues?
>
> Apart from webmail that should be a trivial load.  Webmail systems
> seem to take up lots of resources in my experience, is it an option to
> have a separate machine for webmail?

that machine should have no problems handling that number of users.  in
fact, it's overkill for the job - which is OK, it doesn't hurt to have a
faster machine than you need :)

the biggest load would be, as russell says, the webmail.  fortunately,
you can do a lot to optimise the current setup.

some suggestions:

1. use Maildir rather than mbox if your users are in the habit of
leaving large mailboxes on the server.

you're currently using qpopper, which IIRC copies each mailbox to /tmp
each time it is accessed.  that's bad, very bad - webmail access via pop
means every web-page fetch is causing that to happen. this is almost
certainly the cause of all your load problems.

if you change to Maildir, you'll have to change your pop & imap daemons
to Maildir compatible oneswhich brings us to:

2. replace qpopper with something a bit nicer to your disks.  e.g.
cucipop if you want to stay with mbox format for a minimal and very easy
change. 

or switch to Maildir and install courier-{maildrop,imap,pop} for best
performance.

3. one more thing that will probably reduce the load immediately, even
under your current RH setup is to configure syslog so that mail.log is
written async - i.e. your syslog.conf should say "-/var/log/mail.log"
rather than "/var/log/mail.log".  

4.  also, switch to postfix rather than sendmail.


other semi-random comments:

a nice setup is debian, postfix, courier-maildrop, courier pop & imap,
and maybe courier's sqwebmail (although you may want to stick with
openwebmail as it is a better webmail program).  the only advantage of
sqwebmail is that it interacts with the Maildir files directly, rather
than through imap - this is either an advantage or a disadvantage,
depending on how you want to look at it :)

personally, i don't think it's possible to build a better mail server
than that (debian, postfix, courier etc).  you've already got good
hardware and with that software combination, you'd have best-of-breed
software to matchand it's done entirely with Free Software too.


doing items 1-4 above will probably get the load down enough that your
mgmt may be willing to continue with the linux experiment.

probably the easiest way to do all this is to just rebuild the machine
as a debian box and install the debian packages.



finally, if you use Maildir, i'd recommend using XFS or reiserfs as the
file-system.  ext2 performance really sucks when you have a few
thousands files in one directory, but both XFS and reiserfs handle that
without a problem.

personally, i'd use XFS or reiserfs in preference to ext2 or ext3
anyway.


good luck.

craig

-- 
craig sanders <[EMAIL PROTECTED]>

Fabricati Diem, PVNC.
 -- motto of the Ankh-Morpork City Watch


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]




Re: load average question

2002-11-23 Thread Torsten Krueger
Hi,

On Sat, 23 Nov 2002, Russell Coker wrote:

> On Fri, 22 Nov 2002 23:51, Scott St. John wrote:
> > A few weeks ago we talked about me moving a server from BSDi to Debian.  As
> > luck
> > would have it that BSDi server gave out last Monday and I had to move fast
> > to replace
> > it.  Knowing I can do a RH install in about 30 minutes I went the route of
> > familiar
> > territory and installed 7.2 with Sendmail/QPopper/Apache/OpenWebMail.  I am
> > paying
> > for that now with a huge performance problem.  I am seeing Load Averages
> > spiking
> > above 6 during the day.  Hardware is a Dual P3-600 with a gig of ram on a
> > IBM Netfinity Raid 5 controller.

Hmm - watch out for the load qpopper produces. Since the sendmail/qpopper
combination uses mbox a users mailbox is copied every time the mbox is
accessed. If you have large mailboxes (e.g. no quota an users leave mail
on server) this can produce a significant load. 

> > The owner of the company wants to go back to BSD, but I want to pursue
> > Debian. So the question is:  is anyone running a similar set up with either
> > Sendmail or Posrtfix servicing 2,000+ email accounts with any performance
> > issues?

I would see no problems with that. If you can make sure that the popper
produces the load I'd suggest moving to Maildir for storage of the users
mailboxes and perhaps Postfix with Courier Pop3d.

> Apart from webmail that should be a trivial load.  Webmail systems seem to 
> take up lots of resources in my experience, is it an option to have a 
> separate machine for webmail?
> 
> http://www.coker.com.au/postal/Postal SMTP/POP benchmark
Try this one and look what type of the system is the reason for the
problems. 

Regarding a switch to debian. Three years ago we started switching our
servers from Suse to Debian. I've never regretted that switch since
maintaining Debian is so much easier than Suse due to apt. 

Regards
Torsten Krueger



-- 
Media Online Internet Services & Marketing GmbH
Torsten Krueger   [EMAIL PROTECTED]
fon: 49-231-5575100fax: 49-231-55751098
Ruhrallee 39   D-44137 Dortmund


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]




Re: load average question

2002-11-22 Thread Russell Coker
On Fri, 22 Nov 2002 23:51, Scott St. John wrote:
> A few weeks ago we talked about me moving a server from BSDi to Debian.  As
> luck
> would have it that BSDi server gave out last Monday and I had to move fast
> to replace
> it.  Knowing I can do a RH install in about 30 minutes I went the route of
> familiar
> territory and installed 7.2 with Sendmail/QPopper/Apache/OpenWebMail.  I am
> paying
> for that now with a huge performance problem.  I am seeing Load Averages
> spiking
> above 6 during the day.  Hardware is a Dual P3-600 with a gig of ram on a
> IBM Netfinity Raid 5 controller.
>
> The owner of the company wants to go back to BSD, but I want to pursue
> Debian. So the question is:  is anyone running a similar set up with either
> Sendmail or Posrtfix servicing 2,000+ email accounts with any performance
> issues?

Apart from webmail that should be a trivial load.  Webmail systems seem to 
take up lots of resources in my experience, is it an option to have a 
separate machine for webmail?

-- 
http://www.coker.com.au/selinux/   My NSA Security Enhanced Linux packages
http://www.coker.com.au/bonnie++/  Bonnie++ hard drive benchmark
http://www.coker.com.au/postal/Postal SMTP/POP benchmark
http://www.coker.com.au/~russell/  My home page


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]