Tuning: how distinguish PL/SQL timings from SQL in trace and/or tkprof output?

Question

Hi,
we have a performance problem with one of our customer's databases and are trying to tune it out.
The activity in question is a long-running PL/SQL stored procedure operating more-or-less in batch mode, calling many sub-procedures along the way. The PL/SQL code has been instrumented to take timings of execution of different operations, and we are running with tracing on, and analyzing the trace output using TKPROF.
Oddly, even though we are running through 70+ pages of PL/SQL code, with bulk-collect into large tables and nesting of SQL in other SQL cursor loops, TKPROF is reporting SQL times which account for nearly all the elapsed time reported by our log messages. I mean, we're talking within one or two percent of elapsed time.
How can I distinguish the time spent in PL/SQL operations from the time doing an execute or a fetch in SQL? Should I just believe TKPROF when it says all the time is going into SQL?
Thanks in advance...

Answer 1

A common problem with Oracle timings is the granularity of the clock can give misleading results when viewed over many itterations. That might not be your issue but it's worth noting. I would suggest you use DBMS_PROFILER if you want to get a more accurate picture of your codes performance. In most PL/SQL processes, the SQL will account for the vast majority of the processing time (maybe even 95%+). You really have to be doing some intense string or analytical processing for that percentage to change much.
Richard