Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Usenet -> c.d.o.server -> Re: Performance tuning Oracle Context

Re: Performance tuning Oracle Context

From: Rod Stewart <rod.stewart_at_afp.gov.au>
Date: Wed, 2 Dec 1998 08:29:52 +1000
Message-ID: <741n2d$eb9$1@platinum.sge.net>


What are you retrieving for the results? Are you just retrieving an identifier for each document or the document itself? One of my clients has a similar number of documents, though they may be smaller in size on average. The whole lot takes up about 10 Gbytes of storage. We initially had performance problems (that sound very similar to yours) returning the list of results, which did not include retrieving the actual document. It turned out that disk IO was the problem, to get the 'title' of each document ( which was stored in the same table as the document), it was effectively having to read past each of the long raw documents. We did a little restructuring and indexing so that when we got a document ID from the context index, we were relating it to a document title in a table that did not include the documents themselves. The whole setup is very quick now.

I hope all this makes some sense and is of some use to you.

Rod J. Stewart

rbelanger_at_forrester.com wrote in message <741e16$sdl$1_at_nnrp1.dejanews.com>...
>Hello:
>
>We are working on a Context application that indexes about 1,000 large
>documents (15-18 pages) and about 10,000 small documents (2-3 pages).
>
>Indexing performance seems to be pretty good, but retrieval performance is
>only OK. We are using two-step queries and the average query is taking
about
>10 seconds. I would like to get it down to about 2-3 seconds.
>
>The database server is a Sun Enterprise 3500 with 512 Mb of memory, with
>about a 320 Mb buffer cache, 60 Mb shared pool. We also have SORT_MEMORY
>SIZE turned up to about 8M to keep most sorts in memory. The datafiles are
>on an EMC Symmetrix and are distributed across a series of mirrored
>partitions.
>
>During searches we see a fairly high IO load on the EMC, with almost no CPU
>utilization.
>
>We have tuned our engine preferences to create large tables for the Context
>index tables to prevent chaining and fragmentation.
>
>Any ideas on tuning approaches for Context would be appreciated.
>
>-----------== Posted via Deja News, The Discussion Network ==----------
>http://www.dejanews.com/ Search, Read, Discuss, or Start Your Own
Received on Tue Dec 01 1998 - 16:29:52 CST

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US