We're planning to use JBoss to do clustering for a HA webapp in the near future. The question that I'm toying with is one of tying the JBoss cluster to a database cluster. Our original plan was just to use Oracle RAC, but there's certainly an investment there and it goes up fast if you need more than four sockets. So...
I've been looking at MySQL Cluster (our active dataset is, for the most part, small, and we can use the disk tables for our historical data), and I can't help but wonder. Does this perform well if you have a decent, inexpensive redundant farm for data, and then put a SQL node on each JBoss node, having each JBoss node connect locally to the database? That way we don't need any kind of failover for the DB, per se, and we've reduced two network meshes into one. Admittedly the JBoss/SQL nodes will be busier than they would otherwise be, but that communication pathway is now far faster too.
Is this possible? Inadvisable? Creative? Stupid? I'm sure that I'm not the first one to think of this, but Google and the forum searches in both communities haven't given me a whole lot to go on.
