On Monday,  9 April 2001 at  8:31:40 -0400, Jon Lindgren wrote:
> On Sun, 8 Apr 2001, Kevin P. Neal wrote:
>
> [snip]
>
>> That's better than just hot-swappable, and I wonder if any other
>> system can do the same trick. Can any other processor (Alpha, MIPS,
>> SPARC) failover to another CPU nondisruptively?
>
> Stratus machines did have CPU failover, however, I don't think it
> was to that extent.  IIRC they had 4 processors for each logical,
> and the 4 would constantly check results with each other (via
> hardware... not software...).  When one goes out of service due to
> differences, the other 3 continue on as if nothing happened.  Kind
> of neat; memory was the same way IIRC.  Unfortunately, VOS (its os)
> never realy caught on (too much typing ;-) and FTX (later moved over
> to HP-UX) had it's own set of problems...

IIRC Stratus had 4 processors per CPU, organized as two pairs which
kept checks on each other.  If they disagreed, they both died and let
the other pair get on with it.

Tandem's Integrity series had a thing called triple modular
redundancy: they saved one processor and voted three ways.  Each
processor was on a separate hot-pluggable board, so you could repair
the CPU while it was running.  The Tandem proprietary NonStop
processors were not as sophisticated: they checkpointed data across a
high-speed bus, but they only had a single processor per CPU.  They
were also hot-pluggable.

Greg
--
Finger grog@lemis.com for PGP public key
See complete headers for address and phone numbers