Evan wrote:
> It's my impression that design features such as soldering the controller
> under the mech are considered good practice from a DESIGN FOR
> MANUFACTURABILITY point of view.  And I'll bet that whomever came up with
> that wonderful idea got a bonus for reducing manufacturing costs.
 
It's probably more likely that they did this because it improves the
reliability of the disk.  HP learned long ago that if you solder memory
chips in, rather than putting them in sockets, that the increase in
reliability more than outweighs the requirement that the whole board
be replaced if a single RAM chip fails.  The problem with using this
logic to justify soldering the controller to the disk drive is that
while it probably improves the overall reliability of the drive, it
actually *increases* the catastrophic failure rate, by guaranteeing
that every drive failure is essentally a "mech" failure which requires
the replacement of the part of the drive containing the user's data.
 
As far as I know, HP does not inventory drive mechs and controllers
separately in the field.  When a failure of a controller occurs, the
CE will swap the controller from the replacement drive with that of
the failed drive and then send the whole unit back for repair. So HP
gets back the same number of failed drives regardless of the type of
failure. If the soldered drive<->controller connection improves the
overall reliability of the drive just a little bit, they will get
back a smaller number of drives, making the soldering decision look
like a success.
 
Unfortunately, the number of disk failures that cause the loss of
data may *double* from the customer's point of view (it could be
worse than this because I hear about more controller failures than
mech failures usually).
 
So HP actually sees *fewer* drive failuers, whereas the customer
perceives *more* (catastrophic) drive failures.  A quite interesting
quality problem. :-)
 
G.