You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@spamassassin.apache.org by Robert Menschel <Ro...@Menschel.net> on 2005/02/02 02:17:46 UTC

Re[2]: very slow performance with SA

Hello Alan,

Tuesday, February 1, 2005, 10:59:02 AM, you wrote:

AS> Michael Parker <pa...@pobox.com> wrote:
>>On Tue, Feb 01, 2005 at 10:39:14AM -0800, Alan Shine wrote:
>>>I have one spamd - with the default of 5 max children.

>>This is likely your problem, if you are truly processing at 16 a sec
>>then 5 children probably won't handle the load. Try upping the number
>>of children available. The optimal number is however many you need to
>>handle your peak without overwhelming your system resources.

AS> I understand, but allthough I'm running 5 max children ny CPU is
AS> between 0-2% idle.
AS> (I have dual CPU with hyper thread).

5 max spamd children, 16 emails/sec, means each child is processing
3/sec.  Since you have plenty of CPU available and you are not
swapping, if you double your spamd to 10 children, do you double
your throughput? With or without the SARE rules?

Also, which SARE rules do you have, specifically?  Maybe you can drop
one or two and retain most of the SARE benefit without the performance
hit of the most aggressive/expensive rules files? (Please list ALL
custom rules files, not just the ones with SARE in the file name.)

For that matter, do you also have custom rules?  Any rules in local.cf
or other file?

Bob Menschel




Re: very slow performance with SA

Posted by Theo Van Dinter <fe...@kluge.net>.
On Wed, Feb 02, 2005 at 07:19:46AM -0800, Robert Menschel wrote:
> I can't answer that rate question (is it normal?), since I don't run a
> server, but my mass-check runs process a corpus of 114241 emails
> against all distribution and SARE rules in 10 hours, giving me a rate
> slightly over 3 emails/second on a 2-Ghz 0.5 Gig ram Windows XP machine.

FYI, based on standard 3.1 nightly check (mass-check, standard rules,
no net or bayes), my box (2x 2GHz Opteron) runs 190k messages in 90
minutes, or somewhere around 18msgs/sec/cpu.

-- 
Randomly Generated Tagline:
Stewie: Ah!  Damn it!  I want pancakes.  God!  You people understand
every language except English.  Yo quiero pancakes.  Dali mua pancakes.
Clik clik bloody clik pancakes!
                        - Family Guy, "Love Thy Trophy"

Re[4]: very slow performance with SA

Posted by Robert Menschel <Ro...@Menschel.net>.
Hello Alan,

Wednesday, February 2, 2005, 6:04:56 AM, you wrote:

AS> Hi all,
AS> First of all, thanks for the fast replies.
 
AS> I'll start with the following question: 
AS> 1. Is rate of 16 messages/second normal while running with only rules engine,
AS>    and only installation rules. (DL 380, dual CPU - 2.4GHz,
AS> hyper thread, 4G ram, radhat 8, 10   
AS>    spamd children).

Earlier you had indicated that you were getting this rate with 5 spamd
children. Have you doubled the number of spamd children, and gained no
benefit (and suffered no loss)?

Check out Theo Van Dinter's answer to Pedro Sam in message 23016; are
you maybe unintentionally limiting spamd to using only one child at a
time?

I can't answer that rate question (is it normal?), since I don't run a
server, but my mass-check runs process a corpus of 114241 emails
against all distribution and SARE rules in 10 hours, giving me a rate
slightly over 3 emails/second on a 2-Ghz 0.5 Gig ram Windows XP machine.
(The same mass-check with almost no rules tested runs 10 times
faster.)
 
AS> 2. I tested SA with only one rules file from SARE each time,
AS> non of then is time consuming or   
AS>     loading the system more than the others. 
AS>     When I'm running only with the installation rule sets - the amount of idle CPU is in
AS>      between 10-15%.

If you run top or some similar utility, do all of your spamd children
use up processing time (they should, if they're parallel processing),
or is it just one or a few?
 
AS> 3. these are the sare rules I'm using:
AS> 70_sare_adult, 70_sare_bayes_poisen_nxm,
AS> 70_sare_redirect_post3.0.0, 70_sare_genlsub0, 70_sare_genlsub1,
AS> 70_sare_header0, , 70_sare_header1, 70_sare_highrisk, 70_sare_html0
AS> , 70_sare_html1, 70_sare_oem, 70_sare_random,
AS> 70_sare_specific, 70_sare_spoof, 70_sare_unsub, 70_sare_uri,
AS> 70_sare_top200, 70_sare_bml_post_25x, 99_sare_fraud_post_25x,
AS> evilnumbers.

That set certainly should not be causing any performance problem.

Bob Menschel




Re: Re[2]: very slow performance with SA

Posted by Alan Shine <sh...@yahoo.com>.
Hi all,
First of all, thanks for the fast replies.
 
I'll start with the following question: 
1. Is rate of 16 messages/second normal while running with only rules engine, 
   and only installation rules. (DL 380, dual CPU - 2.4GHz, hyper thread, 4G ram, radhat 8, 10   
   spamd children).
 
and these are the following results to your suggestions:
1. I made sure that I'm running without any network checks, and disabled the URIDNSBL. 
    this didn't help at all.
 
2. I tested SA with only one rules file from SARE each time, non of then is time consuming or   
    loading the system more than the others. 
    When I'm running only with the installation rule sets - the amount of idle CPU is in
     between 10-15%.
 
3. these are the sare rules I'm using:
70_sare_adult, 70_sare_bayes_poisen_nxm, 70_sare_redirect_post3.0.0, 70_sare_genlsub0, 70_sare_genlsub1, 70_sare_header0, , 70_sare_header1, 70_sare_highrisk, 70_sare_html0
, 70_sare_html1, 70_sare_oem, 70_sare_random, 70_sare_specific, 70_sare_spoof, 70_sare_unsub, 70_sare_uri, 70_sare_top200, 70_sare_bml_post_25x, 99_sare_fraud_post_25x, evilnumbers.
 
thanks, for any reply,
Alan.

Robert Menschel <Ro...@Menschel.net> wrote:
Hello Alan,

Tuesday, February 1, 2005, 10:59:02 AM, you wrote:

AS> Michael Parker 
wrote:
>>On Tue, Feb 01, 2005 at 10:39:14AM -0800, Alan Shine wrote:
>>>I have one spamd - with the default of 5 max children.

>>This is likely your problem, if you are truly processing at 16 a sec
>>then 5 children probably won't handle the load. Try upping the number
>>of children available. The optimal number is however many you need to
>>handle your peak without overwhelming your system resources.

AS> I understand, but allthough I'm running 5 max children ny CPU is
AS> between 0-2% idle.
AS> (I have dual CPU with hyper thread).

5 max spamd children, 16 emails/sec, means each child is processing
3/sec. Since you have plenty of CPU available and you are not
swapping, if you double your spamd to 10 children, do you double
your throughput? With or without the SARE rules?

Also, which SARE rules do you have, specifically? Maybe you can drop
one or two and retain most of the SARE benefit without the performance
hit of the most aggressive/expensive rules files? (Please list ALL
custom rules files, not just the ones with SARE in the file name.)

For that matter, do you also have custom rules? Any rules in local.cf
or other file?

Bob Menschel




		
---------------------------------
Do you Yahoo!?
 Meet the all-new My Yahoo! � Try it today!