Oracle FAQ | Your Portal to the Oracle Knowledge Grid |
![]() |
![]() |
Home -> Community -> Usenet -> c.d.o.server -> Re: I/O Performance and Disk Size
<gary_arvan_at_my-deja.com> wrote in message
news:8fhnea$lig$1_at_nnrp1.deja.com...
> Howdy, I'm a new Oracle DBA and I'm already in a debate with my developer
> friends regarding Disk I/O and Disk Size. Customer says the 1 Gig SCSI
drives
> that are 5+ years old have quicker response than the newer and larger hard
> drives of today. He references performance data and magazine articles
that
> are 5 or more years old that clearly show 1gig out performed the 10gig
> drives. But no current data! Consequently, we use 30 refurbished 1 gig
> drives, and yes they do fail rather frequently.
>
> I'm thinking todays 10, 15 or 20 gig drives will out perform these older
> drives, but I have no performance data to back up my claim. Does anybody
out
> there have metrics that I could use to justify my claim?
The german computer magazine c't (http://www.heise.de/ct) publishes regulary
the results of all kind of disks.
Until now I saw steady (but far too slow too keep up with the processors)
increase of
disk performance.
A quote out of there number 9 publication of this year, page 94 of two
current disk models
(not the very fastest may be on the market, but these were in the latests
publication)
Disk name Capacity Rotation CacheRandom Access
Read performance Write performanceHDbench AverageValue
16.1 (min) 23.2 (avrg) 28.6 (max) MByte 15.6 (min) 23.0 (avrg) 29.0 (max) 16.7 15.1 (min) 22.0 (avrg) 25.9 (max) MByte 14.9 (min) 21.9 (avrg) 25.9 (max) 13.7
You could compare that with a dig into my archive of the same publisher a
C'T form June 1995 :
It gives the following benchmark figures for a at the time fast disk
Disk name Capacity Rotation Cache Access Time Seagate Barracuda 4 ST 15150N 4095 MB 7200 960KB 8.0msHD Bech average value Average value
The figures presented here can not compared blindly because of difference in
test-setup during the years at the magazine, the use a now a different
version of the measurement program (hdbench) that publishes more and
figures esp. the IO rate that can be calculated by 1/(random access time in
seconds) .
The Access Time of the old disk is the data as given by the manufacturer.
The access time figure of disk manufactures excluds the time needed for
rotational latency, and the command setup and result processing overhead at
the computer.
What you could read from the above benchmark figures is that the linear read and write rates have gone up by a factor of 5 in the last five year, corresponding with the increase of storage capacity. The random access time has not improved much.
I wouldn't use refurbished disks for a production database system, due to the reliabilty problems you also mentioned esp. if they are repaired disks! Secondhand disk (thus working specimens from another computer) might be o.k. but I would be extremly suspicious if disks have visited a repair shop.
I would use current Ultra Wide (U160 disks), 10.000RPM, 35 GB a piece, and
use also 30 of them.
With U160 you can string 6 of those disks together and you thus need 5 U160
SCSI channels.
You need 6 channels due to the maximum read rate of 30 MBytes/s. 6 Disks
with 6 parallel read requests would saturate a 160 MByte channel.
They would cost about 1000$ a piece and thus you can a have a 1TB storage system for 50.000$ at the moment.
Willem Dekker Received on Sat May 13 2000 - 00:00:00 CDT
![]() |
![]() |