Load balacing spamd

[Paolo Cravero as2594 <pc...@as2594.net>]

Hi.
I have recently setup an SMTP gateway for all the incoming mail to our 
domains with SpamAssassin. We are running two dual-Xeon machines behind 
an hardware load balancer.

Performance is outstanding but training has become a longer process 
since spamd on both machines needs to see what we consider spam/ham, in 
order to keep Bayes and AWL databases aligned.

I wonder if anyone is running SpamAssassin in a load-balanced fashion, 
and how (s)he deals with ham/spam training: RPC, bayes_data brute force 
copy, ... ?

I am still thinking of a solution, so any input will be more than 
appreciated!

Paolo