* Andrew Morton (andrewm@uow.edu.au) [000605 02:09]:
> Tandem had/have a rather different design. It's still single-box
> hardware redundancy, but the CPUs do not run in exact lock-step. The
> CPUs do a periodic rendezvous to compare states; if one is wrong it gets
> ignored/rebooted. The idea here is that if one CPU fails in software
> due to a rare race, the others won't. This might mean that Tandem-aware
> applications have a benefit over naive applications. I forget...
Yes, the CPUs can be individually brought down, if the OS or
software on that CPU cause it to become unstable.
An Non-Stop process (say, a banking transaction program) would have
*two* threads. Only one would be running, the other would be
'pending'. If the running process dies, the backup takes over. The
transaction would be rolled back (via Transaction Monitoring
Facility, aka TMF) and started over.
Controllers can be similarly set up. A connection to a disk array
would look like this:
CPU0 ----------- Controller A -------------- Disk alpha
| \-----\ /------/
| <- inter cpu-bus X
| /-----/ \------\
CPU3 ----------- Controller Z -------------- Disk gamma
the disks are mirrored, and any of these paths, can go down and cpus
can still access the disks. Actually, the inter cpu-bus is a *pair*
of fibre connections that go in a complete loop, so even if *both*
pairs are cut someplace, you can go the other way around the loop to
talk to another CPU.
All in all, pretty clever. But you need to be able to 'hot-swap'
and 'hot-manage' all these devices, from the inter cpu-bus to cpus,
controllers, paths and disks.
Ciao!
-- "So. let me get this straight. You want to fly on a magic carpet to see the King of the Potato People and plead with him for your freedom, and you're telling me you're completely sane?!" --Rimmer (Red Dwarf episode: Quarantine)The Doctor What: Guru to the Gods http://docwhat.gerf.org/ docwhat@gerf.org KF6VNC
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Wed Jun 07 2000 - 21:00:26 EST