Re: OCI performance with user callbacks

roger <rsr_at_rogerware.com> wrote in
news:Xns92FBDA73F3027rsrrogerwarecom_at_204.127.199.17:

> Thanks again.
> I haven't done too much analysis yet, thought I'd
> troll for any easy answers here first:)

FWIW, I don't think there's much expertise on OCI within the NG. I looked for some answers early on and nothing, so that's why I thought I'd respond even though I didn't have an exact match with your situation. Commiserate? :)

> I'll be moving onto 9.2 RSN, maybe things will
> just get better, but I kind of doubt it.

Yah, I kinda doubt it too ...

> I can't imagine why one would use setjmp/longjmp
> to implement invoking some callback hooks, but then I
> suppose it's one way to do it.

Yah, I don't know how it's done that's why I was thinking of a syscall trace to see what exactly is happening. 15% is pretty significant.

> The psuedo code that I saw someplace in the
> oracle documentation (you'll pardon the expression:)
> didn't suggest anything so elaborate was happening.

Okay, so you're finding the documentation as 'lacking' as we found it. Oracle documentation for OCI is very bad and incomplete. ugh.

You said you're on Solaris, if you're using OCIParamGet()/OCI_ATTR_DATA_SIZE the real data type to use is a 'ub2', the main docs state that a 'ub4' ought to be used.

This one through me for a loop for a bit ... our 'shim' compiles on Windows, Solaris, HP-UX, AIX, OSF, on the moon, underwater, etc etc

-- 
Pablo Sanchez, High-Performance Database Engineering
http://www.hpdbe.com