On 26.03.2007 12:35, Robert Klemme wrote:
> On 25.03.2007 03:41, Charles Hooper wrote:
>> A 10046 trace can output wait events that are associated with the
>> individual SQL statements between each COMMIT. This level of detail
>> will likely be more useful to you when performance tuning the
>> application than would the time deltas between the COMMITs. The wait
>> events will tell you, for instance, if the delays are caused by client
>> side processing, writing to the temp tablespace, index or table access
>> (down to the file number and block, which may be used to determine the
>> actual object causing the delay), etc. Each SQL statement will also
>> include the total elapsed processing time and server CPU that was
>> consumed.
>>
>> It takes some practice learning how to read 10046 trace files. Cary
>> Millsap's "Optimizing Oracle Performance" is quite possibly the best
>> resource for decoding 10046 trace files, but there are also a couple
>> good articles on Metalink. While TKPROF can analyze 10046 trace
>> files, much of the useful data contained in the trace file is either
>> lost or reported out of scope. The 10046 trace file may report
>> information like this:
>
> <snip/>
>
> For the purpose of this task (tuning) is there a difference between a
> 10046 trace and DBMS_SESSION.SET_SQL_TRACE / ALTER SESSION SET SQL_TRACE
> = TRUE? I did both in a brief test and it seems that trace file
> contents are mostly similar. Also, TKPROF processes both and seems to
> come up with a similar set of data. Granted I did not do fancy stuff,
> just some selects. Thanks!
PS: On an Oracle 10.2.0.1.0.
Received on Mon Mar 26 2007 - 06:30:20 CDT
