Which number of children spawn is best?

[Monty Ree <ch...@hotmail.com>]

Hello, all.

I have operated some mail dedicated server using redhat 9.0.
About over 4,000 users use this mail server with pop3.
But it is so slow and some spamd seems that spamd can't process well.

So I raise the number of children spawn like below.

SPAMDOPTIONS="-d -c -m9 
==> SPAMDOPTIONS="-d -c -m10 
 
Is it a good method to increase the number to solve the problem?
Or any other good method?

Surely I have read this article.
http://wiki.apache.org/spamassassin/FasterPerformance


# echo $LANG
en_US.iso885915

H/W : Pentium III 1G Dual, 2G RAM, 
OS : Redhat 9.0/ 7.3  / kernel 2.6.11, pamAssassin 3.0.2



Thanks in advance.

_________________________________________________________________
행운의 주인공이 이번엔 나일꺼야, 진짜루... 인터넷 복권   
http://www.msn.co.kr/money/interlotto/  

Re: Which number of children spawn is best?

[Matt Kettler <mk...@comcast.net>]

At 12:44 AM 4/21/2005, Monty Ree wrote:
>I have operated some mail dedicated server using redhat 9.0.
>About over 4,000 users use this mail server with pop3.
>But it is so slow and some spamd seems that spamd can't process well.
>
>So I raise the number of children spawn like below.
>
>SPAMDOPTIONS="-d -c -m9 ==> SPAMDOPTIONS="-d -c -m10
>Is it a good method to increase the number to solve the problem?
>Or any other good method?

Raising the number of spamd can improve overall mail processing thruput, 
however there are some downsides to consider.

1) This will increase memory load on the system. If you add too many 
spamd's the system will start thrashing swap and system performance will 
severely degrade.

2) The increased number of processes can add overhead to the kernel process 
scheduler. With an O(1) (read "order one") scheduler like most recent linux 
kernels use this isn't a big deal, but on some systems with bulky fair 
schedulers this can slow things down if you start using really large 
numbers of spamd's (ie: 100).

3) this will increase load on the CPU, which can make other processes less 
responsive, such as your pop3 daemon. 

Re: Which number of children spawn is best?

[Eugene Kurmanin <ku...@mtmail.ru>]

Hello, Monty.

> Hello, all.

> I have operated some mail dedicated server using redhat 9.0.
> About over 4,000 users use this mail server with pop3.
> But it is so slow and some spamd seems that spamd can't process well.

1. Which MTA do you use?
2. How SpamAssassin is involved in process of scanning?
3. And at what here the pop3?!

> So I raise the number of children spawn like below.

> SPAMDOPTIONS="-d -c -m9 
==>> SPAMDOPTIONS="-d -c -m10 
 
> Is it a good method to increase the number to solve the problem?
> Or any other good method?

1. How much incoming mails do you have per second?
2. How much SA scan the one message at average?

-- 
Kind regards,
Eugene Kurmanin

Re: Which number of children spawn is best?

[JamesDR <ro...@bellsouth.net>]

Monty Ree wrote:
> Hello, all.
> 
> I have operated some mail dedicated server using redhat 9.0.
> About over 4,000 users use this mail server with pop3.
> But it is so slow and some spamd seems that spamd can't process well.
> 
> So I raise the number of children spawn like below.
> 
> SPAMDOPTIONS="-d -c -m9 ==> SPAMDOPTIONS="-d -c -m10
> Is it a good method to increase the number to solve the problem?
> Or any other good method?
> 
> Surely I have read this article.
> http://wiki.apache.org/spamassassin/FasterPerformance
> 
> 
> # echo $LANG
> en_US.iso885915
> 
> H/W : Pentium III 1G Dual, 2G RAM, OS : Redhat 9.0/ 7.3  / kernel 
> 2.6.11, pamAssassin 3.0.2
> 
> 
> 
> Thanks in advance.
> 
> _________________________________________________________________
> Çà¿îÀÇ ÁÖÀΰøÀÌ À̹ø¿£ ³ªÀϲ¨¾ß, ÁøÂ¥·ç... ÀÎÅÍ³Ý º¹±Ç   
> http://www.msn.co.kr/money/interlotto/ 
> 
Depending on how/where sa is called from (remote/local/MTA) may give you 
some options, or limit you in some others.

What I would do if I had the budget (this is my opinion) is to add 
another box, then use round-robin to have the MTA send work to the other 
server(s). This way, you can distribute the load a bit better with out 
raising the children etc.  But this only would work in my case.  I 
currently use the MySQL bayes/awl/user_pref databases. This would scale 
nicely by adding another box with my setup. Another option, if network 
lookups are killing you, is running a local mirror for urirbl/rbl 
lookups. I don't currently have local mirrors, but my mail load isn't 
nearly as high has yours. I've found, in my experience, that sa scales 
better across multiple boxes with a few number of children, than one 
gigantic server with many children. Again, this is my own experience, 
and I can't attest to any others.

-- 
Thanks,
JamesDR