Re: Disk performance for large partitions?

["David Corlette" <corlette_(_at_)_huarp_(_dot_)_harvard_(_dot_)_edu>]

> Yes, there  are quality concerns;  the cheaper boxes are  going to
> fail  more often  and need  more attention. But  they're not  five
> times  more likely  to  fail, even  though they  cost  a fifth  as
> much. I'd argue  that, even  if they're twice  as likely  to fail,
> they  still make  economic sense  at these  prices. My experiences
> are  that  they're nowhere  near  twice  as likely  to  fail. Less
> reliable? Sure, but they're not going to fail twice as fast.  They
> might--*might*--perform  a couple  percent  slower,  but, in  most
> practical situations, they'll still saturate any network link.

In principle, what you're saying makes a certain amount of sense.  My 
experience, however, is that the cheapo boxes don't always work as 
well as you're claiming.  You say, when the first one fails, you just 
drop in the other one.  But what if the second one has the same flaw? 
 And the third?

I had a situation like this with a 2U rackmount from a cheapo 
rackmount vendor.  Since full-height PCI cards didn't fit in the box 
in the normal way, they used a PCI riser card to situate the cards 
sideways.  The PCI riser card went in one of the existing PCI slots, 
and there were little daughter cards that went in the other PCI 
slots.

This was all well and good, but when I put in an Adaptec SCSI card in 
that configuration, things started to fail strangely.  I eventually 
tracked it down to a subtle timing issue relating to the different 
path length in the riser card.  In this case, there was a partial 
workaround, and the system has worked relatively well since then 
(although I lost a PCI slot as a result).  But let's say there wasn't 
a workaround, or the sysman didn't have the tools to debug the 
problem.  So they go and swap out the hardware with a complete 
replacement and lo, it has exactly the same fault!

Compare this with a vendor like Dell or Compaq, who, AFAICT, goes to 
all sorts of trouble to make sure that the physical design of their 
cases et al are fully up to spec.  I very much doubt you'll have the 
same sort of systematic trouble with such a system.

My point being, if you go out and buy three junk computers for $1000, 
you still have $1000 worth of junk.  Whereas if you go and buy a 
properly designed system for $1000, you have a functional system.

Realize that I'm playing devil's advocate here; if you were to see my 
lab, you'd see that I'm running a lot of junk!  But I think that the 
converse point to your argument has to be made and understood.

My 2c, sorry for the OT posting...
                                 Dave C

---------------------------------------------------------
David Corlette          mailto:corlette_(_at_)_huarp_(_dot_)_harvard_(_dot_)_edu
(617)495-5922           http://www-arp.harvard.edu