Re: Warehouse Stripe Size with Oracle8

You don't have to do a very big test to prove the point. If you have some space that you can set up with a few different stripe sizes you need only write a very short C program to do random reads and another to do random writes with the file opened in O_DSYNC mode to emulate Oracle database writes. (I should be loading a suitable pair onto my website in the next couple of weeks).

Unless open VMS somehow does something amazingly different from Unix, your intuition is correct; 512 bytes is daft. Use raw devices and go for a stripe size that is at least the same as the block size, and if you expect to be doing largely lots of tablescans and sorts, then one good rule of thumb is 1 or 2 x the size of the maximum single I/O.

Jonathan Lewis
Yet another Oracle-related web site: www.jlcomp.demon.co.uk

howard.galusha_at_thehartford.com wrote in message <7abuhg$mn4$1_at_nnrp1.dejanews.com>...
>I am a data warehousing consultant working with Oracle8. I have some
>experience with Oracle on UNIX, but I'm certainly not a DBA and I'm looking
>for advice with the following issue:
>
>We are running Oracle 8.03 on a 4-way DEC Alpha under VMS (please don't ask
>why, it is a long story.) We have over 300 GB of disks, set up in various
>RAID arrays (RAID 1, RAID 0+1, RAID5).
>
>Our system DBA's set up the Oracle blocksize to be 32K (to oltp folks, that
>may sound somewhat high, but for data warehousing, it's considered kosher.)
>The disk arrays were set up with a 512 byte stripe size.
>
>Now, intuitively, the stripe size sounds way too small - I picture the o/s
>getting swamped performing 64 i/o's to read 1 Oracle block (and in data
>warehousing, you are reading lots of blocks).
Received on Wed Feb 17 1999 - 15:31:00 CST