<br><br><div><span class="gmail_quote">On 3/15/07, <b class="gmail_sendername"><a href="mailto:matthew.garman@gmail.com">matthew.garman@gmail.com</a></b> <<a href="mailto:matthew.garman@gmail.com">matthew.garman@gmail.com
</a>> wrote:</span><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">On Thu, Mar 15, 2007 at 11:49:56AM +0000, Stroller wrote:<br>> The quote "we observe little difference in replacement rates
<br>> between SCSI, FC and SATA drives" is from a paper which "looked<br>> at 100,000 drives, including HPC clusters at Los Alamos and the<br>> Pittsburgh Supercomputer Center, as well as several unnamed
<br>> internet services providers". If these are not environments able<br>> to stress hard-drives, then I'd be interested to hear of<br>> environments that do.<br><br>Long-range artillery tests? :)<br>
<br>My take on this is that perhaps, at one point in time, it made sense<br>for high-end drives (SCSI, etc) to use larger bearings, thicker<br>platters, etc. But the mainstream hard drive market has existed now<br>for what? At least 20 years. So I'm sure design and manufacturing
<br>has improved, even on commodity drives.<br><br>But I'm sure hard drive manufacturers make a small fortune on their<br>"enterprise" class drives, so they *need* to be able to<br>differentiate them somehow.
<br><br>I think it's akin to selling a passenger car with a frame built for<br>a tank: *sounds* impressive, certainly costs more, but even under<br>stressful use, a car's frame should be the last component to fail.
<br><br>A more precise reason why there's a *visible* difference between<br>SCSI and SATA/PATA drives is probably the MTBF (mean-time between<br>failures). Google or wikipedia MTBF; for hard drives, it is derived<br>
from the MTBFs of all its *components*. (The exact derivation I<br>don't recall.) Point is, bigger/beefier components probably have<br>bigger MTBF ratings. But (1) by its own definition, a big MTBF does<br>not imply that any one drive will last a long time and (2) the study
<br>quoted earlier supports the idea that SCSI's bigger/better<br>components and MTBF ratings don't mean anything in the "real world",<br>even under "enterprise class" use.<br><br>FWIW, Google also published a similar study (I think it's safe to
<br>assume they have *lots* of hard drives and push them all pretty<br>hard).<br><br>I like this solution to the hard drive reliability problem: use the<br>least cost drives possible, but always have three live, up-to-date
<br>copies of your data. (Of course, what you save in cheaper drives<br>you probably end up paying in electricity, cooling and environmental<br>impact!)<br><br>-Matt</blockquote><div><br> </div><br></div>What are you people arguing about. It is well known that the extra price of a SCSI drive is due to the
<span style="font-weight: bold;">longer warranty</span> not due to vastly better hardware. Anybody who uses the same harddrives (in an array or not) for several years knows that they fail. The difference is with IDE/SATA you are usually paying out of pocket for the replacement because the warranty has expired. If you are planing on upgrading your harddrive(s) in the next three years or so go with the cheap drives. If you are buying the drives to use for longer then SCSI's warranty is usually worth the extra price.
<br>-- <br>_____________<br>Ryan Patterson