Large pool usage and advanced queueing

Question

Hello All,
I'm using Oracle 10.2.0.3 under Linux IA 32Bits.
I've got a big problem with my large pool that keep increasing as I'm using Advance queuing. After a while, errors 4031 are raised.
Can someone explain how to monitor large pool usage. Espaecially I would like to know wich process/ session consumes large pool.
Thanks in advance for your help

Answer 1

Whilst Daniel does his 'Twenty Questions' trick on you (he has a valid point: OS and Oracle Version are ALWAYS useful to know... but there's no need to pin it down to four decimal places for such a simple question!), I'll have a go at actually providing an answer.
A number of things might be going on.
First, do a google search on "joxs heap init" and you'll find suggestions that setting or increasing the JAVA_POOL_SIZE can fix the problem. See this for example, which has exactly your problem:
http://www.mail-archive.com/[email protected]/msg36154.html
Second, a lot of things that would use the Large Pool if it exists can instead make use of the shared pool if it doesn't -but they fragment the thing mightily as they do (the Large Pool was invented precisely to provide a chunk of memory that doesn't age out, only gets re-used and thus never fragments). So something (an RMAN backup, for example) might work without a large pool, if there's room in the shared pool; but if the shared pool is already fragmented or tight, then it will error with a 4031. Allocate it a bit of Large Pool, however, and it prefers to use that, finds the memory it needs is now available and therefore works fine.
The large pool is used for RMAN, shared server, I/O slaves and parallel slaves. Who knows precisely what's going on in an export that might be touching on at least 2 of those things without you explicitly asking for it to do so?! I mean that it's possible it's doing a bit of 'internal parallelism' to make itself efficient, for example, and failing to find the memory it needs for inter-parallel-slave communications. Just a possibility and a bit of speculation on my part.
I suppose the next experiment would be: take Large Pool back down to zero and then set Java Pool to 15M instead (note you should really be allocating in units of 4M these days, though Oracle will round up anyway).