Ron Seybold wrote after I wrote:
 
>> Does this mean that 2GB drives will be dying a slow death -- or that they
>> will not be obtainable much longer?  Frankly, principally because of I/O
>> throughput considerations, 2GB is about as much as I'm willing to put on
a
>> spindle.
 
>What kind of improvement do you need in I/O transfer rates to justify a
>bigger device than 2Gb per spindle?
 
Well, if I've got 2 X 2GB disc drives which can each do, say, 30 I/O's per
second and can each transfer data at, say, 10MB/sec, then, a 4GB disc
replacing them would have to be capable of 60 I/O's per second and a
20MB/sec transfer rate to give me the same throughput.  (note that these are
arbitrary performance numbers).
 
<snip>
 
>Ken Paul also said that AutoRAID is very big in HP's futures for
>3000-level storage devices. If one of these fault-tolerant RAID units is
>made of 10 2Gb-devices (or hopefully, more reliable ones), is it a 2Gb
>unit anymore? Does the redundancy improve the value beyond 2Gb reliability
>rates, and at what kind of cost?
 
For me, the jury is still out on RAID with respect to performance.  When I
first started reading about RAID a few years ago it occurred to me that,
while RAID would protect data integrity, it would do so at the cost of
performance -- because while in a non-RAID environment 8GB might have been
on 4 spindles (each with a disc controller) and two channels, post-RAID it
would be 5 spindles (with one controller for all five mechs) on one channel:
a reduction of, say 120 aggregate I/O's per second down to 30 I/O's per
second.  Some of my friends didn't see that connection, converted to RAID 5
arrays and subsequently got to deal with performance problems.  In fact,
 I'm still hearing war stories about sites that went RAID and saw their disc
performance fall through the floor -- I recall that there was a posting on
3000-L within the past week or two...
 
 
 -- Evan