🚀 go-pugleaf

Re sustained transfer rates on workstation HDDs:
 (and bypassing the interface discussions).

    Anyone care to justify why they are always talking as though
 the main activity of a workstation was peeling data off
 contiguous sectors/tracks/cylinders and belting it down the bus?
 Certainly there are applications in which this may be true;
 perhaps video.  But in a very great deal of disk-intensive
 workstation activity, the actual data-flows are much more
 complicated, even with a single active process.  Consider the
 updating run of an accounts or data-base package, where there will
 be quite a large number of files open and being accessed in a
 pretty random manner.  In such a case the disk-heads will be flying
 about all over the place, and the limiting factor is much more
 likely to be the latency of getting the head to the right track
 from wherever it was left and waiting for the sector to come round
 than peeling off many Kb of data.  And this is a very real-world
 situation; updating runs of this sort are some of the slower
 activities in commercial computing.

    Added factors to be considered:  (1) for important data,
 read-after-write will probably be switched on;  (2) often there
 will be more than one process active, with further randomization
 of access;  (3) even the most-often-defragged disks will have
 fragmented files simply because applications make and delete them
 very freely (and defragging takes valuable time).

    There's probably a practical seat-of-the-pants formula for this
 sort of activity, giving the relative importance of burst and
 sustained transfer speeds, average seek/search, and the effect of
 large-scale buffering.  But only the practical experts will know
 what it is.

    Of course, it's not as much fun thinking about this as arguing
 about the relative merits of your own pet hardware.  But then,
 where would hardware development get to if there weren't a body
 of people who were mad keen to push it to the limit?  It's like
 the fact that my Ford or Fiat does actually benefit from the
 activities in Formula I.

                                        Chris K.
-- 
Chris Kennington (newshound)                        01491 574065
  You can mail me with suitable modifications to header-addresses,
   or to "chris <at> inf-eden.demon.co.uk", replacing the <at>.