Berkeley db, how to tune it ?

[Gilles Hamel <ha...@laposte.net>]

Hello,

Last weeks, amount of received spam increased fastly. Now, our spamassassin server is becomming I/O bound to manage bayes and whitelist berkeley db. Spamassassin wiki suggests to stop using bayes, but it is not an acceptable solution here.
I have seen some people who fix this issue in putting db berkley in ramdisk and copy it every day on persistant storage. Other people talk about a dbd4 shared memory buffer pool. How to use it ? There is no information about it in wiki or faq. Documentation talks about a DB_CONFIG file with set_cachesize parameter, but to use it application must create a db environment ... 

http://pybsddb.sourceforge.net/ref/env/intro.html

Thank you

Re: Berkeley db, how to tune it ?

[Theo Van Dinter <fe...@apache.org>]

On Mon, Dec 04, 2006 at 09:48:08AM +0100, Gilles Hamel wrote:
> Last weeks, amount of received spam increased fastly. Now, our spamassassin server is becomming I/O bound to manage bayes and whitelist berkeley db. Spamassassin wiki suggests to stop using bayes, but it is not an acceptable solution here.

I don't think there's a lot of optimizations that you can do to BDB.  You may
want to consider switching to SQL.

I'm not sure why the wiki suggests turning bayes off.  Perhaps it's suggesting
disabling auto-expire, since that tends to be the big i/o suck.

-- 
Randomly Selected Tagline:
"Apres moe le deluge" - Larry and Curly get wet