Re: spamprobe-1.1x7 released

On Wed, 30 Mar 2005 10:32:24 -0500
Brian Burton <[email protected]> wrote:

> Jacob S wrote:
> > On Wed, 30 Mar 2005 10:12:53 -0500
> > Brian Burton <[email protected]> wrote:
> >>The record size in a hash database should be the same on 64 bit 
> >>processors as it is on 32 bit processors.  I haven't tested that
> >>though  since I don't have a 64 bit system to test on.
> >>
> >>A great test to run would be to build a hash database on a 32 bit 
> >>system, export that to a file, import that file into a database on a
> >>64  bit system, export on that 64 bit system to a new file and then
> >>diff the  two export files.  They should be identical.
> > 
> > hmm... I was thinking previously (with PBL) the size difference was
> > only noticed when doing a du/ls -l on the db. When exported to a
> > text file, both dbs were the same size. Is there something different
> > about hashing that I don't know about, or is this a case where I
> > don't know anything and should keep quiet? :-) 
<snip>
> The hash file code assumes that records have a fixed size independent
> of  word size on the machine SP is running on.  I think I got that
> code  right but I can't test it here.  The test I outlined was to
> verify that  I don't have a bug which might corrupt the hash file on
> 64 bit CPUs.

Ok, I ran some tests to see how spamprobe compares on a 32 bit system
and a 64 bit system. The 32bit system in my tests was an AMD Athlon 1800
and the 64bit system was a 440Mhz SparcIIi in a Sun Ultra 10. Both
systems are running Debian. Perhaps also of note is that the Sparc64
port of Debian uses a 64bit kernel with a 32bit userland (which of
course means some libraries are installed with both a 64bit and 32bit
version). I did not supply any special flags to spamprobe when
compiling, so I do not know how it compiled.

The spam bank used to build the DBs was in the form of 6 mh mailboxes.
Three were ham (18.2MB), three were spam (108MB). 

A hash db exported from the 64bit and imported on the 32bit before being
re-exported returned the exact same size when doing an ls -l. The same
was true when a hash db was built on the 32 bit system, imported on the
64 bit and re-exported. 

However, building a hash DB directly from the spam bank on each system,
exporting it and running an ls -l on them showed a 44.4KB difference
between the two. The 32bit hash export was the smaller of the two. The
differences all seemed to be of the following form: 

2,0,960,"I0xed54fbe0"
0,1,960,"I0x96551cfa"

HTH,
Jacob


-------------------------------------------------------
This SF.net email is sponsored by Demarc:
A global provider of Threat Management Solutions.
Download our HomeAdmin security software for free today!
http://www.demarc.com/info/Sentarus/hamr30