On Mon, 5 Jun 2000, David L. Nicol wrote:
> Bruce Guenter wrote:
> >
> > On Sat, Jun 03, 2000 at 02:29:34PM +0100, James Sutherland wrote:
> > > This is, IMHO, quite an attractive idea: a fully hot-swappable system,
> > > where any failed component can be replaced without any downtime.
> >
> > And how do you plan on swapping out the motherboard that everything
> > connects into?
>
> Migrate all user processes onto a different box,
> shut down,
> repair or replace,
> bring up,
> rejoin cluster.
Hrm. The current box is dead. There's nothing to migrate; at best, you
could pull back a previous snapshot of the system state.
> Given migratable tcp-ip connections
> http://tcp-impl.grc.nasa.gov/tcp-impl/list/archive/0696.html
> this could be done transparently.
>
> Without, another host could set up a bochs/vmware like
> environment to contain everything the computer that is
> in need of repair is running, and it would be activated at some
> "instant" the same time the NIC on the host host starts
> listening on the broken host's IP address and routing those into the VM.
At the very least, each machine needs enough fault tolerance to keep going
long enough to transfer processes and load to a replacement. I suspect
there will be some performance hit while migrating the load from one
machine to another with one machine functionally impaired? This, of
course, assumes that all the systems DO have enough redundancy to survive
any single component failure...
Otherwise, you are just looking at "Oh dear, the server just died. Here's
a replacement we can install instead", but faster.
James.
-
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 - 21:00:20 EST