Re: [APC-DEV] APC b0rkage upcoming

Rasmus Lerdorf wrote:
> I know we have talked about it privately, but it might be a good idea to 
> give folks on this list a summary of the goals of the pooling mechanism.

Yup, the pool idea has been in my head for quite a while now. It's there
to avoid a lock call for every allocation in the system.

Initially, it was supposed to be per-process, but it has been reworked
to be per-cache entry. Every entry holds a pool and allocates all memory
required from the pool, which pulls in 4/8k blocks. This should speed up
the allocation operation, but that's not the key problem I'm trying to
fix.

More importantly, most segvs in APC seem to happen during cache full
conditions when the cache expunge is being performed. Having a pool
essentially reduces all free calls into just deallocating the pool
blocks (i.e 1+4 sma_free calls for 16k of memory).

It'll also reduce the cruft of the cleanup code whenever there's an
allocation failure and indirectly ensure that there are no memory
leaks.

Hope all that made sense.

Cheers,
Gopal
-- 
If I learn from my mistakes, pretty soon I'll know everything