On 10/18/2010 12:54 PM, Török Edwin wrote:

> What kind of signatures do those 3rdparty databases have?
> Can you use wc -l, and then group them by extension?
> I would expect hashes (.mdb, .hdb) to load quite fast, since we have
> lots of those too, and .ndb, or maybe .ldb to load a bit more slowly if
> you have many.

# wc -l *
    2310 bytecode.cld
  140903 daily.cld
  705232 main.cld
  ------
  848445 (67%)

     139 sanesecurity.ftm

      20 sigwhitelist.ign2

     111 doppelstern.hdb
    1613 rogue.hdb
     854 spamimg.hdb
      63 spamattach.hdb
    2158 winnow.attachments.hdb
   14344 winnow_malware.hdb
   -----
   19143 (2%)

      56 spam.ldb
       3 winnow.complex.patterns.ldb
    ----
      59

  216439 INetMsg-SpamDomains-2m.ndb
     542 doppelstern.ndb
   36235 junk.ndb
   19492 jurlbl.ndb
   49131 jurlbla.ndb
    2217 lott.ndb
    1727 mbl.ndb
   14604 phish.ndb
   11167 scam.ndb
   20878 scamnailer.ndb
   15439 spear.ndb
    3943 spearl.ndb
    1901 winnow_malware_links.ndb
     709 winnow_phish_complete_url.ndb
  ------
  394424 (31%)

> Although it is expected that using more signatures slows down DB
> reload, 4 minutes is a bit much.
> How long does it take with only the official DBs?

I don't know, I'll have to test that on the off-hours.

BTW clamd spikes the CPU to between 80 to 90% during those 4 minutes.

> That should actually speed up the DB load, since it doesn't have to JIT
> compile the code for Sparc.
> You only get slowed down during a scan.

OK, so no improvements will come from that (I just changed compilers to
gcc 4.4.5 and compiled as 64-bit to see if anything improved, nothing
really changed from the old gcc 3.4.6 and 32-bit; other than I don't get
the warnings about g++ being too old and file descriptors too few).

> AFAIK it supports it only for static code generation (i.e. it requires
> an assembler).

OK.
-- 
René Berber


_______________________________________________
Help us build a comprehensive ClamAV guide: visit http://wiki.clamav.net
http://www.clamav.net/support/ml

Reply via email to