[Scottish] Drive performance...

Andrew Back andy at smokebelch.org
Fri Dec 1 08:38:02 GMT 2006


On Fri, 1 Dec 2006, Kyle Gordon wrote:

> Cure my curiosity folks... before I open up the machine and put the kit in for
> some bonnie++ tests...
>
> What is likely to be faster... both hosted in a Dell 1600SC and set up in a
> software (mdadm) RAID5 setup. Total space is not an issue.
>
> 3 18.2GB Quantum Atlas 3 drives, connected to an Adaptec 29160 - giving 36GB
>
> or
>
> Caviar 80GB (Master) + DiamondMax 60 (slave) on the internal IDE controller +
> Deskstar 120GXP on a PCI Ultra100 TX2 card - giving 120GB

What configuration gives 120GB from 80 + 60?

> Am I wrong in thinking that the SCSI drives will be faster than the IDE setup,
> given their age? The DiamondMax is ATA66 whereas the Caviar and DeskStar are
> ATA100, but the Quantums are Ultra160. The Quantums however, are older, and
> the primary use of this will be ~, where random access will be preferred over
> sequential streaming.

It may not be the case any more given advances in IDE/ATA technology but 
it certainly used to be that all other things being equal SCSI would win 
where the workload was of a more random nature. It's bus protocol is 
(was?) more advanced and allowed command queueing. The OS could send 
a bunch of requests at the drive and it would be able to service them in 
the order it saw fit based on where the heads where at. Whereas with IDE 
everything was serialised and the drive would have to wait for the blocks 
to pass the heads, service that request, and then take another request, 
wait for the data to go by the heads and so on.. So SCSI made sense in 
file servers and multiuser systems, and IDE in the likes of a video 
editing workstation where access would be largely sequential.

Of course then you have to factor in drive the performance, cache and so 
on. And overhead/benefits of disk configuration options - RAID*/JBOD.

There may be other benefits to SCSI I've missed, and I admittedly know 
little if anything about modern ATA drives.

Andrew



More information about the Scottish mailing list