[plug] [OT] opinions: SCSI hard disks

Bernd Felsche bernie at innovative.iinet.net.au
Sat Apr 26 10:05:29 WST 2003


On Sat, Apr 26, 2003 at 03:26:37AM +0800, Craig Foster wrote:
> Jon Miller <mailto:jlmiller at mmtnetworks.com.au> wrote:
> > If you look on the IBM web site and look up Linux and benchmarking you
> > will see their benchmark test using rhl7.3 and their server with all
> > the details of what they did.  As for drives since I use IBM SCSI
> > drives all the time, I've yet run into any problems that I keep
> > hearing about on any of the servers that I've built.  I would
> > recommend that you look at the 36 or 73GB HDD.  As for the onboard
> > controller, use it only for the backup (all onboard SCSI controllers
> > are usually (90%) of the time a strip-down version of the rreal deal.
> > I tend to buy an separate card from either IBM or Adaptec for the
> > drives. 

I've found them acceptably-fast on the IBM x235 and similar.

What you have to keep in mind is that each 15 krpm drive can deliver
in excess of 60MB/second as a sustained rate so plugging more than
4 into a U320 channel is likely to produce contention if access to
the drives is randomly distributed and applications are
data-intensive. Contention is possible with just two drives on a
channel, but that seldom presents a problem (streaming many
concurrent broadcast-quality video could well be an issue). 60%
bus-loading is about the maximum you should start with, until the
data access profile is well-defined.

If data access requirements are "bursty" but small in total volume,
then you can put more fast drives on the same SCSI channel.

> Search through the PLUG archives for AD (or active-directory - can't
> remember which), propagation, and RAID.
> 
> Those two dead RAID5 members were IBM 36GB USCSIs. 
> Two drives, not a week old from the wholesaler, died at the same time
> (10min difference)

Check the manufacturing dat and batch number. If two drives from the
same batch fail, it means that there's either a serious production
issue (very unlikely), or that the batch has been mis-handled on the
way to you.

OTOH, you may simply not have read the installation guidelines and
complied with cooling or electrical requirements.

> No more IBM SCSI for me :/

I had a Seagate die and wedge a whole SCSI bus for me yesterday.
No more Seagates for me.... maybe I should mention that the drive is
about 8 years old. The drive's older cousins in the same system have
however not failed yet.

-- 
/"\ Bernd Felsche - Innovative Reckoning, Perth, Western Australia
\ /  ASCII ribbon campaign | I'm a .signature virus!
 X   against HTML mail     | Copy me into your ~/.signature
/ \  and postings          | to help me spread!



More information about the plug mailing list