RE: bottleneck analsyis on spamassassin

2006-07-26 Thread Chris Santerre
Title: RE: bottleneck analsyis on spamassassin







 -Original Message-
 From: Ramprasad [mailto:[EMAIL PROTECTED]]
 Sent: Wednesday, July 26, 2006 6:54 AM
 To: spamassassin-users
 Subject: bottleneck analsyis on spamassassin
 
 
 Hi,
 Spamassassin has so many dependencies on various external factors 
 like network , disck IO , RAM etc 
 If I want to analyse the performance on my SA box , how do I find out
 what the bottlenecks are. 
 I am using spamassassin as a module in Mailscanner on CentOS
 Is there any tool by which I can analyze the bottlenecks of my system


I can tell you right now, its either Net tests or poorly written rules. Otherwise SA runs pretty darn good. 


Spammassassin -D --lint  some_test_email


--Chris 





RE: bottleneck analsyis on spamassassin

2006-07-26 Thread Ramprasad

 I can tell you right now, its either Net tests or poorly written
 rules. Otherwise SA runs pretty darn good. 
 

Darn good is how good ? 
On a Dual Xeon with 4GB ram can SA scan 30k mails per hour. 
Today at 15k the machine starts signalling problems , 20k is the max it
can do beyond which there are unacceptable delays 

 Spammassassin -D --lint  some_test_email
 
How do I know what percentage of time is taken by individual tests ? 

Thanks
Ram



Re: bottleneck analsyis on spamassassin

2006-07-26 Thread Dhawal Doshy

Ramprasad wrote:

I can tell you right now, its either Net tests or poorly written
rules. Otherwise SA runs pretty darn good. 



Darn good is how good ? 
On a Dual Xeon with 4GB ram can SA scan 30k mails per hour. 
Today at 15k the machine starts signalling problems , 20k is the max it
can do beyond which there are unacceptable delays 


Spammassassin -D --lint  some_test_email

How do I know what percentage of time is taken by individual tests ? 


Try installing mailwatch in a limited manner (mailwatch.sf.net, works 
only with mailscanner), you do not have to use it for quarantine 
management or any other thing.. under tools you have a link for a lint 
test which gives you nicely formatted html page with the time taken for 
each test.


As Chris mentioned.. network tests (especially razor/pyzor) could be 
responsible for SA delay. If you have a lot of servers try running a 
local only pyzord (see pyzor.sf.net).


- dhawal


Thanks
Ram


Re: bottleneck analsyis on spamassassin

2006-07-26 Thread Theo Van Dinter
On Wed, Jul 26, 2006 at 08:38:31PM +0530, Dhawal Doshy wrote:
 How do I know what percentage of time is taken by individual tests ? 

You can't really, without profiling the perl directly.

 As Chris mentioned.. network tests (especially razor/pyzor) could be 
 responsible for SA delay. If you have a lot of servers try running a 
 local only pyzord (see pyzor.sf.net).

Yeah.  At that point it's a question of waiting as opposed to resource usage.
If you disable network tests, you'll find CPU is the main bottleneck, followed
by RAM.

-- 
Randomly Generated Tagline:
Make like a bottom and split.


pgpsu2ZJpRR6L.pgp
Description: PGP signature


Re: bottleneck analsyis on spamassassin

2006-07-26 Thread jdow

top

Network checks are costly for time. Ping times can give useful hints.

Some rule sets are extraordinarily costly for memory and time. (There
is an evil rule set that's reasonable and another that is obsolete,
overgrown, and a huge system drain. Be careful to pick the right one.
Read the descriptions on RulesEmporium. {^_-}

If top shows your system is thrashing (has actually used some of the
swap space) you either have too many children active or too little
memory.

With multiple children disk IO is not too likely to be a problem with
modern disks using spamd or anything else which demonizes spamassassin.
Even with only one child it should not be noticeable compared to the
rest of the time spent actually scanning and on network tests. This
of course changes if you get into using the swap file. That is the single
most time expensive thing that can happen to SpamAssassin other than a
lost network path to the DNS block list lookups or broken DNS on your
system. (And a ten second DNS timeout is probably a tossup with SA going
into swapping heavily with a lot of children.)

{^_^}
- Original Message - 
From: Ramprasad [EMAIL PROTECTED]




Hi,
 Spamassassin has so many dependencies on various external factors 
like network , disck IO , RAM etc 
If I want to analyse the performance on my SA box , how do I find out
what the bottlenecks are. 
I am using spamassassin as a module in Mailscanner on  CentOS

Is there any tool by which I can analyze the bottlenecks of my system

Thanks
Ram