[lug] RH Linux download + hdparm

[D. Stimits]

John Karns wrote:
> 
> On Tue, 23 Oct 2001, D. Stimits said:
> 
> > John Karns wrote:
> 
> > > Thanx for the response.  But I'm not too sure about the gains.  I'm
> > > running a U160 drive with an ultra-wide (40 MB/s Tekram 390U) controller
> > > on an Asus P5A mobo (33 Mhz PCI, 500Mhz K6-II cpu).  hdparm says I'm
> > > getting 15 MB/s.  I had been wanting to order a U160 controller to replace
> > > the 390, but under the circomstances, I'm afraid that it could be a waste
> > > of $$.  Perhaps I'll try a quick benchmark with an Adaptec 2940UW for
> > > comparison to see if it's a controller issue.
> >
> > If your drive only runs 15 MB/s, you wouldn't have much to gain. The
> > exception is if your PCI bus is already bandwidth challenged by some
> > form of i/o, then the burst rate would be very helpful. SCSI, unlike
> > IDE, is rather intelligent, and you don't generally worry about things
> > like DMA settings. On the actual SCSI cable, devices can get
> > instructions and then detach until they have something to send back
> > (other drives get to work). Thus drive cache ram that can burst quickly,
> > and generally small data transfers as well, will grow in benefit as you
> > add more scsi drives to a single cable. On the PCI bus end, it has to
> > operate at the bus speed, the question then is how well your specific
> > controller cooperates with other devices.
> 
> I'm mystified.  The drive is a 68-pin Quantum Atlas V with active
> termination, rated at 29MB/s, 7200rpm.  Must be the controller or mobo.
> The machine is not process bound either (virtually idle), so I doubt that
> the bus is i/o bound.  Although it could be a mobo design issue I suppose
> - it's showing its age.  It would be interesting to read some performance
> specs on those mobos.

What is the total capacity? I don't think there is such a thing as a
7200 rpm drive that can average 29 MB/s. They lied. Unless maybe it is
one of those 180 GB drives. You'll get 31 MB/s with a 10k rpm 36 GB
drive with reasonably dense clusters. Are you sure there is no fine
print with some stipulation like "29 MB/s on outer tracks only, and
unfragmented systems"? Sorry, I'm skeptical of any 7200 rpm drive that
claims over 20 MB/s.

> 
> > Now if you know for a fact that you are going to use scsi, and buy new
> > drives in the future, you might still want to get U160. I say this
> > because although controller cards are significantly more expensive, the
> > U160 drives of otherwise similar spec to uw or u2w are only a few
> > dollars different (and many of the lower end drives are going away).
> > Once you absorb the controller cost, drive expense by interface isn't
> > that different. And you will never find an uw 10k rpm or 15k rpm, they
> > will never be built.
> 
> I've always been sold on SCSI - not for i/o speed so much as the gain in
> multitasking performance.  I will probably upgrade the mobo before looking
> for a U160 controller though.

You never need to worry about dma settings or cpu useage on scsi. Not in
all cases, but in many, it plays nicer with the system as well. As soon
as you add multiple devices to it, advantages show up real fast in
comparison to IDE.

D. Stimits, stimits at idcomm.com

> 
> ----------------------------------------------------------------
> John Karns                                        jkarns at csd.net
> 
> _______________________________________________
> Web Page:  http://lug.boulder.co.us
> Mailing List: http://lists.lug.boulder.co.us/mailman/listinfo/lug