Oracle FAQ | Your Portal to the Oracle Knowledge Grid |
![]() |
![]() |
Home -> Community -> Mailing Lists -> Oracle-L -> Re: IO rate for U320 SCSI 15K RPM Hard Drive
Agreed.
In fact, it is not uncommon -- when price is no object -- to use the outer
10% or less or the spindle(s). I base this
remark mostly on things like TPC benchmark full disclosure reports, where
the discerning (and maybe even
not-so-discerning) reader will note that the disk space provisioned for the
database is 10x (or more) than what
is actually used. You can be certain that the benchmarkers are not buying
all of those disks and leaving 90%
of them untouched...
The advantages of using only the "outer X%" of a disk are pretty well
known. If you choose X low enough
(usually doesn't need to be much lower than 10, I would think) you should be
able to drive your "average"
seek times amazingly low. And it doesn't exactly hurt the throughput for
(large) sequential I/Os, either.
Sadly, my personal experience differs considerably from Alex's; I have very
rarely been able to convince
management that a RAID-5 stripe of 10x500GB ATA disks might not be ideal for
high demand databases,
or that the corporate ERP databases perhaps should not share physical
spindles with the e-mail system.
Selling managers who are fixated on metrics like dollars-per-gigabyte (or
dollars-per-terabyte now) on such
concepts can be very difficult. ;-)
In defence of the dollars-fixated managers, though, this sort of
consideration is probably only relevant for the
largest and most demanding of environments. With suffiently large block
cache in the database (for reads)
and NV cache in the disk array (for writes), it is not too terribly
difficult to engineer a (modest) database that
very rarely performs physical I/O... Well, except at backup time. Or
reporting. or...
Don't get me wrong -- I am a firm believer in BAARF, and I really do think
building a 4TB database on 6 disks
is almost always insane. But I am also starting to encounter more and more
databases where the memory
(available to be) used for block buffers ranges from 60% to 200% (!) of the
total database size. (Yes, I
have encountered servers with 8GB of RAM and 2GB databases...)
On 10/13/06, Alex Gorbachev <gorbyx_at_gmail.com> wrote:
>
> At one company I worked for we divided the disk into 4 quarters and
> used only outside two or one for most demanding databases. In extream
> cases we could go as low as 12 or less percent of disk surface.
>
> ...
-- Cheers, -- Mark Brinsmead Senior DBA, The Pythian Group http://www.pythian.com/blogs -- http://www.freelists.org/webpage/oracle-lReceived on Sat Oct 14 2006 - 15:51:15 CDT
![]() |
![]() |