Michael Katz a écrit : > Stan Hoeppner wrote: >> Eero Volotinen put forth on 11/30/2009 2:14 AM: >>> Quoting Ali Majdzadeh <ali.majdza...@gmail.com>: >>> >>>> Stan, >>>> Hi >>>> Thanks for your detailed response. Actually, the main reason which >>>> drove us >>>> toward performing virus scanning as an offline process was >>>> performance. As >>>> we deal with large amounts of e-mails, we found the way amavisd-new or >>>> other >>>> filtering management tools performing filtering too slow. We >>>> intended to >>>> somehow decrease the amount of load which amavisd-new or similar tools >>>> impose on the architecture. > > > There are many filtering Postfix AV solutions that are far more > efficient than Amavisd and many AV scanners that are considerably more > scalable than clamav such.
I'd be happy to see more arguments about this. and please don't tell me "perl is slow" or the like. I'd like to see more quantitative measurements (to see which parts need to be improved). > A few years ago we did some detailed testing > between ClamAV a few years ago, clamav was indeed very "slow". but since then (one year ago? I don't remember), it progressed. did you redo your tests lately? and what does the clamav tests have to do with amavisd-new? did you measure amavisd-new? if so, how? (yes, this is an open question, not a provocative one). > and commercial av scanners and the difference was huge in > terms of load reduction and throughput. In our tests we have found that > the biggest performance limitation in Postfix for AV/AS scanning, > assuming you have removed bottlenecks that amavisd and clamav introduce, > is from having to copy messages out of the queue to scan. Some > commercial email platforms allow for scanning in memory rather than > requiring copying files and these platforms , in our test, far outscale > Postfix for filtering over a 100 messages/second. >
