bah - Team Stratus!!
None of this wanna be software redundancy - real computers do it hardware !!
/jk
Always was a bit of rivalry between them (at least with Wall St. data centers)
/jk
Always was a bit of rivalry between them (at least with Wall St. data centers)
When they still had their own multi-board proprietary processors, Tandem also used lock-stepped microprocessors in their maintenance and diagnostic subsytem and some of their communications controllers.
I also gather Stratus is still around running on x86-64.
It seems like it would be an enormous challenge to keep modern CPUs exactly in lockstep. All of the same interrupts have to occur at exactly the same times and be handled in the same amount of time? (I guess the hardware has to physically ensure that every interrupt is raised on both CPUs at the same time?)
They used a custom chipset that compared the two MIPS processors and also synchronized them on external interrupts. I believe that Tandem System Review is on archive.org. Volume 8, Number 1 (Spring 92) has info on the first generation lock-stepped TNS/R, and Volume 10, Number 1 (January 94) has info on the 2nd generation MIPS "Himalaya K1000".
For Itanium, the processors were loosely synchronized, but not really lockstepped. They ran the same application code, but the processors could have different cache, TLB hits, etc. They got synchronized on external IO, all of which went over the ServerNet. There is a paper called NonStop® Advanced Architecture[0] by David Bernick, et. al that has the details on their 1st generation Itanium systems.
[0] https://www.cse.chalmers.se/edu/year/2012/course/ZZ_courses_...
(the only source I could find that had the paper)
Stratus was pretty nice - I liked VOS :-)