Re: Process Migration on Linux - Impossible?
Darren Reed (darrenr@cyber.com.au)
Thu, 2 Oct 1997 15:11:55 +1000 (EST)
In some mail I received from Victor Yodaiken, sie wrote
[...]
> So, I don't see why, if we are interested in load balancing
> or fault-tolerance, or distributed tasks, we should immediately start
> by assuming that process migration is the first requirement. Maybe for
> compute intensive components of a program we can get the programmers
> to identify compute blocks that have simpler state and are easier to
> move around than "processes", but that can really benefit from being
> run on a lightly loaded processor. Maybe we can figure out an extension to
> BSD ^Z signal that would package up program state in such a way that
> a new executable is created that can be restarted later. Maybe
> a "process" should be a traveling token that drops off work at net sites
> and constructs a data flow machine on the fly. Maybe gui processes
> and servers and computational processes are all very different kinds
> of things that should be treated differently by the os. ...
what about thread migration ? that could be an entire process or just a
thread.