Re: Table performance problems with RAW.

Gary Eckhardt wrote in message <6trml6$lps$1_at_client3.news.psi.net>...
>Hello all.
>
>We have come across an interesting problem that I thought I would ask the
>rest of the world if they've seen the same thing.
>
>Our current setup is Oracle 7.3.4 Parallel Server running on AIX 4.2. The
>performance problem comes when we have tables with a "medium to large"
>initial extent (50M) and are performing unindexed full table scans of the
>tables, such as in a SELECT COUNT(*) FROM EMP. These selects are taking a
>very long time on EMPTY TABLES. I have not seen this behavior in the
>"cooked" file system world, and about the only thing I can attribute it to
>is running on RAW file systems. The larger the initial extent, the longer
>the select time and vice-versa.
>
>Currently we've gone through our code and tried to find all the full table
>scans and make them indexed reads, and have shrunk down the size of some
>tables where we cannot force an indexed scan to get around the situation.
>However, I believe there must be some type of RAW tuning parameter
somewhere
>that will alleviate some of our problems.
>
>Any suggestions/comments about this?
>

My opinion is that your problem isn't in the consequence of using raw devices.
I have no experience with AIX but nothing similar has been happening to me with other UNIXes. Never heard about any raw tuning parameter. Really, there is nothing to tune with raw devices. There is nothing but device driver  peace of
code ) between your disks and RDBMS - no buffers, no lists. How long does it really take to complete query ? As far as I know, in your case RDBMS scans through all 50 MB. It happenes regardless your database files are on "cooked" or "raw" devices.

Good luck ! Received on Mon Sep 21 1998 - 04:52:35 CDT