On Sat, 02 Dec 2006 18:31:47 +0100, Noc Phibee <[EMAIL PROTECTED]> wrote:
>Thanks to your answer > >Yes 6 server in load balancing with for all 70 concurrency incoming >only for spam detect and 3 server for virus scan > > > > > >Michael Scheidell a écrit : >>> -----Original Message----- >>> From: Noc Phibee [mailto:[EMAIL PROTECTED] >>> Sent: Saturday, December 02, 2006 8:35 AM >>> To: users@spamassassin.apache.org >>> Subject: Best Choice for Bayes filtering on SpamAssassin >>> >>> >>> Hi >>> >>> i have 6 servers running on spamassassin 3.1.7 (now after a >>> upgrades). Actually, all have Bayes Filering with local Db >>> (default db, not sql) >>> >>> I want know what is the best choice ? : >>> - Default Db or MySQL db ? >>> >> >> MySql. Db can corrupt. Db WILL corrupt, and you can't replicate/share >> it. >> >> >>> - 1 Bayes Db per server or 1 Bayes on Sql for all server (same >>> database) >>> >> >> If all 'round robin' mx, maybe 1 bayes on each server. >> >> If priority (mx 1, mx 2, mx... Etc) having one bayes per server will >> give a very jaded view of the world for server 6 (spammers go for >> highest # mx first. All it will get is spam) >> >> Also depends on why 6 servers, are all 6 the same? Load balancing? >> Failover? Backup mx's? different functions? >> (some do SA, some do cached dns, some do mysql, some do postfix?) >> >> >> >> >>> My server receive 500 000/ 750 000 mails /days >>> >> >> I have one getting 10MM per day. >> >> Configured right, you would really only need two servers, the other 4 >> make an update/configuration problem. >> >> With 3 servers, you could try mysql nbd database (I have not yet tried >> this) >> With 2, you could try mysql replication dual-master/slave(and deal with >> collisions, collision skips might not be a big deal) >> >> With 2, you might try memory devices, and 'mirror' the memory device >> which would hold the mysql server (I have not tried this, I don't think >> that a missing record or two on the bayes db is any bid deal) >> >> You COULD, once per day, just after expire, dump/load the Bayesian from >> 'master' to slave. >> >> >> Hi, I run multiple SA server fronts end with a single MySQL bayes backend and have done for a number of years. At some point I'll add Load Balancing to the SQL but at the moment it's on a stable box with little or nothing else to do. To date I've had no issue with it; though my mail throughput is a fraction of yours. The reason for multiple SA's is/was to cover downtime on any given server for maintenance. My MTA has a list of SA servers it will use in series; if 1 is unavailable it will got to 2 and so on. How this would work under the heavy loads you experience is open to debate. All I can say is that it's worked very well here. HTH Nigel
