Phil,

Is the source, less OCO modules, for CP still distributed with VM? I wonder
if the CP maclibs or source for z/VM would have a MVCL burdied somewhere in
them following a lock of some kind.

Paul 

MVCL and CAS are for different purposes, even though they do similar things.
MVCL us used to move data around (and it is actually a copy since the
original data is duplicated).  CAS is used to synchronize multiple
processors.  The trick with CAS is that it is autonomous, which means
locking out any other processors while the compare and swap take place.
This usually means that some kind of hardware lockout is put in place before
the value is fetched from memory, then releasing the lock after the new
value is stored back.

The only thing I can think of that is related to reliability is that a MOVCL
will not work for multiprocessor synchronization since it will not prevent
two processor from modifying the same memory at the same time, while the CAS
will.

-----Original Message-----
Phil,

By themsleves the cycle times might not be of interest. The cycle times in
conjunction with explanation of fetch operations and how instructions and
their addresses are really resolved would probably be very instructive.

Not complex examples but a few that illustrate why a compare and swap is
more reliable in a multi processor environment and an MVCL isn't.

Paul

-----Original Message-----
.
IBM does publish cycle times in some of their "in depth" discussion of
zSeries processor designs.  However, there are so many other factors
involved in zSeries performance that these are almost an aside in the
discussion of the design.  I have a list of cycle times for the last twenty
or so mainframes announced including Amdahl and Hitachi products - I've
never felt it worthwhile to publish them because they don't really add a
great deal to our knowledge and they distract from the real factors.

--
  Phil Payne
  http://www.isham-research.com
  +44 7785 302 803

Reply via email to