[Info-vax] VMS port to x86

[John Wallace]

On Mar 24, 7:46 pm, JF Mezei <jfmezei.spam... at vaxination.ca> wrote:
> John Wallace wrote:
> > Tandem hasn't needed CPU lockstep since they dropped their own
> > proprietary processor architectures,
>
> But building a fault tolerant system is much more than "lockstep". There
> are many hardware designs (dual power, dual fans, dual disk controllers
> etc etc).
>
> Are you saying that NSK now runs on the same systems/model numbers as
> HP-UX and VMS ?

As John Reagan already said, not on the same HP model numbers as VMS
and HP/UX boxes, but on the same CPU chips. Contrary to some writeups
elsewhere in the past, no "lockstep CPUs" or lockstep memory are
needed - not at instruction level anyway.

The fault tolerance in modern NSK (ie in NonStop Advanced Architecture
systems) is provided largely by "logical synchonisation units" which
compare the results of chunks of calculations before they are written
to the outside world (disk, network, etc); they do not compare the
timing and results of individual instructions (that would not be
sensible in modern hardware).

Obviously that is a massive oversimplification. For more detail, see
what you can find to helpfully read about NonStop Advanced
Architecture and (as mentioned in my earlier post) Logical
Synchronisation Units.

As also mentioned in my earlier post, the Hal Massey (VP of Tandem)
presentation to OzTUG in 2006 which was the best I had found on the
subject, sadly seems to have vanished, I didn't keep a copy, and I
haven't found anything nearly as helpful anywhere else. Suggestions
for substitutes most welcome.