dspam data sizes

[Doug Hardie]

On Mar 18, 2005, at 02:50, Christian Damm wrote:

>
> this was discussed in detail (several times) on the dspam mailing list 
> - search the archives: 
> 'http://dspam.nuclearelephant.com/search.shtml'.
>
> you can extremely cut down your db size using different 
> methods...which training method you are currently using? - once dspam 
> is "mature" enough (training wise) it is (for example) a good idea to 
> switch to TOE mode (train on error) if your user base is large - your 
> db shrinks and the performance gain is big.
>
> Doug Hardie schrieb:
>> After seeing the recommendations here I am testing dspam.  I 
>> currently have 2 users testing it.  They are at both extreme ends of 
>> mail demands.  My account gets about 1000 emails a day of which about 
>> 25-30% are spam.  The other account gets about 100 emails a day of 
>> which about 90% are spam.  So far, dspam is quarantining about 1/3rd 
>> of the spam I receive.  Is basically the same as Apple's mail 
>> filtering.  My mail is being sent to both.  However, the storage of 
>> the data used by dspam is a bit overwhealming.  The storage is at 250 
>> MB for me and 20 MB for the other account.  I have cut way down the 
>> purge retention intervals which appears to help somewhat but not 
>> enough.  I have thousands of users and at 250 MB per user (it would 
>> probably be a bit below that)  its just not practical.  How do other 
>> users of dspam deal with this issue?  Is there some setting I have 
>> wrong (or at least not set most efficently)?

Thanks.  I dug through the archives and it appears the recommend 
approach is to use BDB4 as the backend with TOE mode, chained disabled, 
and frequent dspam_clean -p0, and -s0 purges.  My current setup is:

sqlite backend
TOE mode
nightly purges using the provided script but with the days reduced to 3.
Will have about 3K users when in production
chained has been disabled
I believe all other settings are default except that I have to use Home 
directories for the data storage.