Re: Reasonable performance ?

In message <1158306954.533175.308790_at_i42g2000cwa.googlegroups.com>, "frank.van.bortel_at_gmail.com" <frank.van.bortel_at_gmail.com> writes
>
>Adrian Simpson schreef:
>
>> We're not entirely sure that we have got a system that is performing
>> well.
>>
>> We had a partitioned table with 2254 partitions, of which 260 had data
>> in, varying between 1 and ~41,000 rows.
>
>> The DB is set up to parallelise using 16 processors, but at any one
>> time, only 8 processes (ora_pxxx_<dbname>) seem to be active.
>>
>
>41k Rows is not much - I would say the overhead for the parallel query
>is more than the time it would take to retrieve these.
>Although I am not absolutely certain your queries run prallel 8,
>I would suggest to take a closer look at your break even point.
>Relatively few records are processed faster without parallel query.
>What actually is "few" is entirely for you to find out.
>My experience is the tbreak even point is
>at about 200,000 rows. Different hardware, different setup, so
>not comparable at all!
>

The table was only just starting to be populated. In the longer run, we would expect 40,000+ rows/partition, but querying could be over several partitions. Basically it is a partition per day, and we could be querying over 3 months, so at that we are getting up to a substantial number of rows. Whether or not at 16 the parallelisation is right for this sort of figure I'm not sure, but we did notice an improvement elsewhere in the DB when we upped it from 8.

Something else to have a fiddle with by the looks of it.

Thanks

Adrian

-- 
To Reply :
replace "bulleid" with "adrian" - all mail to bulleid is rejected
Sorry for the rigmarole, If I want spam, I'll go to the shops
Every time someone says "I don't believe in trolls", another one dies.