Re: vxfs and Oracle db_file_multiblock_read_count

what kind of hint or help do you need ?

HP-UX always performs i/o on vxfs file systems in file system blocks. it appears that your file systems were created with an 8K block size.

direct i/o provides a way to bypass the file system buffers, which avoids the normal overhead of a memory to memory copy that would otherwise be performed. direct i/o does not alter the file system block size.

oracle may indeed be performing 256K read operations (assuming that the segment has extents >= 256K). but the oracle read will be translated, by the file system, into a series of file system block size (8K) i/o operations. at the os level, this will be translated into i/o operations of a multiple of the os block size (512 bytes)

with your file system block size set at 8K, i'm assuming that the database block size has been set to 8K as well. having your file system block size the same as the database block size is about as good as you're going to get with file systems.

you do realize, of course, that setting multiblock read count to a high value is going to tell the optimizer to assign a lower cost to full table scans...

HTH "Luis Santos" <lsantos_at_pobox.com> wrote in message news:9611gb$pr$1_at_nnrp1.deja.com...
> Hi!
>
> We use Oracle 8 (8.0.6.2.0) and HP-UX 11. All our Oracle fs vxfs
> file systems.
>
> I discovered using the tusc utility that HP-UX is doing 8K reads on
> datafiles, regardless the Oracle db_file_multiblock_read_count
> parameter, which is tuned to do 256K reads.
>
> We read that we must use the direct_io options of vxfs, mounting it
> with the parameter convosync=direct
>
> Well, we did everything right but the reads are still been done in
> 8K chunks.
>
> Any hint or help is welcome!
>
> --
> Luis Santos
> Oracle DBA
>
>
> Sent via Deja.com
> http://www.deja.com/
>
Received on Sat Feb 10 2001 - 17:01:05 CST