Well, i just find why the import is so slow. Oracle8i linked against glibc2.1 is
3 times slower than Oracle 8.0.5 with glibc2.0; so the RAID0 has no influence
because the time of reference is for Oracle 8.0.5+glibc2.0. I have installed the
old version of Oracle and then build the RAID0, and now is about 25% faster!
Thanks everybody.
Eladio Linares Morcillo wrote:
> O.K., I was trying to understand why the performance for Oracle was poor
> with RAID0. With no raid and 8 files 2GB each, across 4 SCSI disks and two
> scsi hosts; an import of 1GB + analyze runs for 4 hours. With raid0 i killed
> the process after 8 hours.
> The machine is a HP LH4, two Xeon 400Mhz, with two ncr53c8xx scsi hosts
> and 6 UW2 disks of 9GB each, kernel 2.2.5 and raidtools 0.9. The raid0 is
> composed by 5 full disks, and i tryed chunk-sizes 4K,16K,32K,128k and 256K,
> mke2fs -b 4096 -R stride=(chunk-size/4096); but performance for import was
> always poor.
> any ideas?
> Thanks, Eladio.
>
> Eric Ladner wrote:
>
> > The db_block_size is not the size that oracle reads or writes in the
> > manner that you are thinking. From the OS level, that is still based on
> > the IO routines of the host operating system.
> >
> > The db_block_size is the minimum allocation size on the MEDIA. i.e. if
> > you have a device that the block size is 4K, you don't want to put 2K
> > blocks on it, but 4K instead.
> >
> > This parameter is usually OS driven and typically doesn't have much to do
> > with the database at all. On Linux, I'd pick 8K as a decent size...
> >
> > Eric
> >
