Re: speed questions
Date: 1996/10/25
Message-ID: <54prgp$7po_at_cf01>#1/1
"David M. Taulbee" <dtaulbee_at_po.michsb.trw.com> wrote:
>Steve Rentz wrote:
>> We have a 50 gig database in Oracle 7.1 (currently testing 7.3). Raw
>> data consists of about 15 gig. We have the data denormalized to avoid
>> joins. The amount of time it takes to get results from queries we feel
>> is unacceptable (3 hours for an average query, 8 hours for anything
>> complex). The database consists of 3 main tables (customers - 13 million
>> records, orders - 27 million records, items - 80 million records). We
>> are wondering if going to a star schema would improve performance? Any
>> recommendations would be appreciated.
>In the past, it has seemed to me that denormalization has rarely been of
>much benefit. We have a small system that has one table with 40+
>columns and only 20,000 rows that requires substantial more time to
>produce simple queries and updates than a NORMALIZED system consisting
>of 66 tables of which 1 table alone requires 100+ Meg of actual data
>storage (this is smaller than its actual database size).
>The keys here are normalization, indexing and TUNE YOUR QUERIES. You
>also need to use EXPLAIN PLAN to verify that your queries are using your
>indexes.
>David M. Taulbee
>TRW Vehicle Safety Systems
>dtaulbee_at_po.michsb.trw.com
Denormalisation is quite a big problem.
We have many time optimisation and speed up some application
with denormalisation.
But denorrmalisation have a cost. We have to maintain coherence between
normalised and duplicate information.
So we often use denormalisation to avoid systematic joins to get only
one information or to get result of fonction ( max, avg ... ).
After denormalisation you have also to tune your database and queries. Remenber that the writing of the query is the most important to get good response time.
Alain Chereau
As usual :
statements and opinions are mine and do not necessarily reflect the opinions of my employer. Received on Fri Oct 25 1996 - 00:00:00 CEST