Article 172003 of comp.os.vms:
By an interesting coincidence, the current issue of the IBM Systems
Journal discusses Sysplex technology.  My local Borders actually had a
copy on the racks.  I didn't feel like spending $18 on it, so only
glanced through it.  A quick read indicated that, indeed, the latest
versions of Sysplex support many of the things VMS cluster users have
come to know and rely on over the years:  True global locking,
distributed I/O, etc.  I believe they now support up to 32 processors in
a Sysplex.

It's been clear for years now that IBM designers have been following the
VMS cluster technology closely.  Sysplexes go back some 8 years or so, I
think, and have followed an evolutionary path somewhat like VMS
clusters.  The initial releases were vaguely like VMS's original
pre-cluster support for CI devices.  But they've gotten much more
sophisticated over the years.

The focus of Sysplexes and VMS clusters has been different, however. 
For example, Sysplexes had explicit support for database transactions
before they had general-purpose locking; VMS clusters built database
transactions on top of a general-purpose lock manager.  Sysplexes
supported very long-distance connections quite early on; I recall
someone at IBM telling me about a Sysplex with nodes in New York and
Japan at least 4-5 years ago (though this was an internal setup - I have
no idea if it was possible for customers to build such aggressive
configurations).

Sysplexes seem to have focused on performance and disaster tolerance. 
To this end, Sysplex designers have been willing to include special
support for (and in) high-use products (like DBMS's) and to require
specialized hardware.  VMS clusters started out by requiring specialized
hardware (CI's), but soon evolved to support any network connection with
adequate performance.  (They'll use specialized hardware like CI or
Memory Channel when it's available, of course.)  While DEC, when it
still owned RDB, explicitly modified it to make heavy use of cluster
features, most software that's really cluster-cognizant is part of the
VMS infrastructure (e.g., file system code, shadowing); the goal is that
most applications simply run transparently.

VMS clusters don't have the hardware support necessary to match what a
current, high-end Sysplex can do in terms of performance.  However, the
VMS lock manager is much more general (if I understood the papers I
scanned through quickly, the Sysplex lock manager has just shared and
exclusive modes).  In general, there are many commonalities, and many
differences, mainly due to different requirements and constraints.

In comparison to either VMS clusters or Sysplexes, most of what passes
for "clustering" out there corresponds to technology VMS and Sysplex
users saw in initial releases many years ago.

							-- Jerry