Of clustering patterns and JBoss
Cache help is at hand...
Editors' Blog It sometimes seems that everything lives in cache these days and I worry about the complexity of interacting cached systems built on hardware clusters.
It could be all too easy to optimise throughput for a particular job mix and hardware configuration; such that if any of the assumptions changes slightly, throughput drops through the floor. What you need is broad "satisfying" throughput plateaus, not sharp optimising peaks.
And then there are possible "emergent behaviours" from large numbers of concurrent, interacting, cached, transactions in what is, essentially, a nondeterministic environment....
I was talking about this area with Bela Ban, lead JGroups manager JBoss Clustering Group, at JBoss World, about the need to, perhaps, refactor caching products so as to reduce user choice. A bad decision on what to cache and how to configure the cache may be obviously catastrophic in production, but it may not be equally obvious beforehand.
Another approach is to make proven patterns (with prerequisites and consequences/Use Case) available, of course. And Bela has just pointed me at the clustering patterns  page at JBoss.org. It is, he says, "just in the starting phase, the goal is to compile a list of patterns/best use practices". But, if you're into caching systems running on clusters - and who isn't these days - it is well worth watching.
It's not clear how well these patterns will generalise, they are stated for JBoss cache, but I feel there must be a generic pattern behind them... ®