Here is more information for ya'.
Yes, use replication. However, clustering, and what you're prolly getting at is load balancing, is very difficult among multiple database systems. There are people that do it, but I don't trust it. There is more information on the web concerning the related issues.
Here is what I do. (Yes, I am responsible for such a system, btw). Load balance things like web servers or transaction servers and run the database machines in a fail over mode. In that way, you have one db machine acting as the primary. Information from that primary machine is replicated to other backup db machines.
As for clustering software, there are a number of options. First make sure you know exactly what you mean when you say the word cluster. Beowulf does clustering, but treats all machines in the cluster as one machine, as opposed to load balancing machines. BIG difference. If you want to load balance web servers, ftp servers, mail servers, and things like it, Pirahna would do fine and is distrubed with Red Hat.
Now if you are like me where you have to services that are load balanced and a third that is fail over (the databases) AND you also want a way of monitoring not only the health of hardware, but also software (the server software and so on), AND support multiple platforms, then I would suggest TurboCluster from TurboLinux.
I should still have a flow chart of the network design. Let me know if you want to see it.
Later on,
Big Din K.R.
