TreeCache tuning question

[newton]

I want to cache a lot of objects of the same type (currently 250,000, probably approaching a million in a year or two). As I understand the way treecache works, if there are any changes on a given node, the entire contents of the node are replicated. If that's true, I probably need to distribute my objects around the tree so that smaller batches are replicated with every update.

So I was thinking about using the key of each object to contruct a more unique cache region. Say if the object's key was "seattle", I'd store it in the cache region /s/e/a/t or /s/e/a/t/t/l/e or something.

The question: With TreeCache, is it more efficient to have a really deep tree with very few objects on each node, or a flatter tree with more objects on each node, and what are the tradeoffs?

M.

[ben.wang]

No, only the (key, value) pair gets replicated not the whole node.

-Ben

[newton]

OK, that's good to know. So is there any benefit in making a deep tree like I am doing, or should I put all 250000 objects on one node? Which is better, and why?

M.

[belaban]

The locking granularity is the *node*, so if you only have 1 node (e.g. /myRoot), then different threads will lock on /myRoot. Of course, if you only have 1 threads, that doesn't matter