On 27/01/2010 22:08, Stevan Bajić wrote:
> On Wed, 27 Jan 2010 18:42:05 +0100
> "[email protected]" <[email protected]> wrote:
>
> On 27/01/2010 18:04, Stevan Baji wrote:
> >>> On Wed, 27 Jan 2010 17:32:18 +0100 "[email protected]"
> >>> <[email protected]> wrote:
> >>>
> >>>> no problem, i will be pleased to give some feedback. Actually i'm
> >>>> running a french email service provider company and all ours
> >>>> filtering servers working with dspam.
> >>>>
> >>>> Migration to 3.9.0 was very easy and no problem was reported.
> >>>>
> >>> Phuu... that was important to me personally that the update is
> >>> easy.
> >>>
> >>>
> >>>> We've see many improvements in processing speed, that is very
> >>>> important to us.
> >>>>
> >>> For us too :)
> >>>
> >>>
> >>>> Our filtering setup using: amavisd 2.6.4 clamav with extra sig SA
> >>>> 330 Dspam 3.9.0 CRM114
> >>>>
> >>>> if CRM114 rely on a SA plugin to work and report result, SA and
> >>>> DSPAM are natively called by amavisd. Until last week, DSPAM was
> >>>> called in command line mode (amavisd default setup), but now
> >>>> we're using daemon mode. so amavisd setup is now the following:
> >>>>
> >>>> @spam_scanners = ( ['SpamAssassin',
> >>>> 'Amavis::SpamControl::SpamAssassin'], ['DSPAM',
> >>>> 'Amavis::SpamControl::ExtProg', $dspam, [ qw(--client --stdout
> >>>> --process --deliver=innocent,spam --user), $daemon_user ],
> >>>> score_factor => 1, ], );
> >>>>
> >>>> We have not yet see a great improvement in speed or mysql load,
> >>>> but i think it's better this way.
> >>>>
> >>> Speed should be slightly better when using daemon mode. You could
> >>> patch Amavisd-New to use dspamc instead of dspam binary. Then you
> >>> would probably see an additional speed imporvement. But not ultra
> >>> spectacular. Just a bunch of seconds less per day used when calling
> >>> DSPAM.
> >>>
> >>>
> i would give it a try
> >>>> With this setting we can generate dspam signature, so we can
> >>>> easily retrain on false positive (cron script)
> >>>>
> >>>> The only problem is that we are not able yet to autolearn on
> >>>> mistake, like CRM114 can do as a SA plugin, according to total
> >>>> score.
> >>>>
> >>> Aha. To get thet you would need to add DSPAM as an SA plugin.
> >>> CRM114 is AFAIK directly integrated into SA. Right? And DSPAM you
> >>> call from within Amavisd-New. Right?
>
> indeed
> i use this SA plugin for CRM114:
> http://mschuette.name/files/crm114-0.8.0-sa33.pm.
> Did a DSPAM plugin will be ok for performance or it's a bad choice ?
>
> > From the performance viewpoint not much would change (if at all) if
> you would integrate DSPAM as an SA plugin. But the
> possibilites/flexibility would be greater.
Ok, so I think i need to look for someone who can make this plugin.
>
>
> >>> What CRM114 setup do you have? What classifier? I assume you use
> >>> OSB Winnow? With unique microgrooming or just microgrooming or
> >>> plain OSB? What SPAM/HAM threshold do you use? Do you use a
> >>> symetric or asymetric threshold? Do you call directly CRM114 or do
> >>> you use mailreaver or mailfilter or any other CRM114 script? Which
> >>> one?
> >>>
> >>>
> CRM114: version 20090807-BlameThorstenAndJenny (TRE 0.7.6 (BSD))
>
> the SA plugin call CRM114 like this:
> /usr/bin/crm -u /var/amavis/.crm114 mailreaver.crm
>
> > Okay. You are using mailreaver.
>
>
> to be honest, i've used the default settings for crm114 so:
> :clf: /osb unique microgroom/
>
> > OSB. The same as DSPAM has except that you have "unique" which DSPAM
> does not implement. The microgroom is anyway not needed in DSPAM.
>
>
> :good_threshold: /10.0/
> :spam_threshold: /-5.0/
>
> for threshold, i suppose SA plugin change crm114 default settings with
> these: (into crm114.cf)
> crm114_staticscore_good -1.0
> crm114_staticscore_prob_good -0.5
> crm114_staticscore_unsure 0.0
> crm114_staticscore_prob_spam 0.5
> crm114_staticscore_spam 3.0
>
> > No. That's used for SA's internal scoring and has nothing to do with
> the above mentioned CRM114 threshold.
>
>
> # custom crm114 thresholds
> # default values are +/- 10
> crm114_good_threshold 5
> crm114_spam_threshold -5
>
> > That's another story. Where do you have those values? In the
> Amavisd-New configuration? You should change that to match the values
> mentioned in mailfilter.cf
>
these are modified values from SA plugin crm114 from
http://mschuette.name/files/crm114.cf
I will change these to match mailfilter.cf
>
> the only problem with CRM114 is that css files are local on file
> system, so we cant share between multiple servers like mysql backend
> in SA or DSPAM.
>
> > You could with NFS or any other network based FS, but it's a PITA.
>
> > And don't forget that DSPAM has stuff like group support that CRM114
> does not have. It is possible to do merged groups in CRM114 but out of
> the box none of the script in CRM114 has it. Let alone innoculation
> and classification groups. On the other hand CRM114 has way, way, way
> more classifiers then DSPAM and even has neural networks which DSPAM
> does not have.
>
>
> >>>> Our most busy filtering server actually manage around 80 000
> >>>> emails per day (effectively processed into amavisd), the others
> >>>> are closed in volume. Average scan timing are around 800 ms on a
> >>>> Intel Bi Xeon E5530, 24 GB RAM, SAS disks. SA takes around 80-90%
> >>>> of this time.
> >>>>
> >>> It is no surprise that SA is using the bigger part of the
> >>> processing time.
> >>>
> >>>
> >>>> DSPAM is the fastest with an average scanning time of 60 ms !
> >>>>
> >>> :) For DSPAM I usually messure how many mails per second DSPAM can
> >>> process and for SA I messure how many seconds per mail SA needs.
> >>> Huge difference IMHO.
> >>>
> >>>
> >>>> Load average is around 0.4
> >>>>
> >>>> 2 SQL servers (Master/Master) are external to filtering server. 7
> >>>> filtering servers (with dspam) are connected to these sql
> >>>> server. We've have no replication problem at all with dspam even
> >>>> in master/master.
> >>>>
> >>> I have the same. So far no issues except that I need to take care
> >>> where to run the purge script since both masters should not do the
> >>> purging at the same time. The new purge script does proper locking
> >>> so even if you would run the purgin on both masters, only one will
> >>> purge a table at the same time.
> >>>
> >>>
> >>>> Voila ! Thanks for all the works !
> >>>>
> >>> Thanks for the verbose feedback.
> >>>
> >>>
> you're welcome
> >>>> Regards Tonio
> >>>>
>
>>
>>
------------------------------------------------------------------------------
The Planet: dedicated and managed hosting, cloud storage, colocation
Stay online with enterprise data centers and the best network in the
business
Choose flexible plans and management services without long-term contracts
Personal 24x7 support from experience hosting pros just a phone call away.
http://p.sf.net/sfu/theplanet-com
_______________________________________________
Dspam-user mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/dspam-user
> ------------------------------------------------------------------------------
> The Planet: dedicated and managed hosting, cloud storage, colocation
> Stay online with enterprise data centers and the best network in the
business
> Choose flexible plans and management services without long-term contracts
> Personal 24x7 support from experience hosting pros just a phone call away.
> http://p.sf.net/sfu/theplanet-com
> _______________________________________________
> Dspam-user mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/dspam-user
------------------------------------------------------------------------------
The Planet: dedicated and managed hosting, cloud storage, colocation
Stay online with enterprise data centers and the best network in the business
Choose flexible plans and management services without long-term contracts
Personal 24x7 support from experience hosting pros just a phone call away.
http://p.sf.net/sfu/theplanet-com
_______________________________________________
Dspam-user mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/dspam-user
