Question: How is circuit-switched routing actually implemented? It seems really hard to have probing happen both in parallel and concurrently with other packets coming through the network. Do you associate a lock with each node in the network and just try to take locks along the path? This seems like it would be wrought with inefficiency.
This comment was marked helpful 0 times.
dtaveras
@wcrichto I'm not sure what you mean by packets coming through the network conflicting with probing because if its a circuit switched network all packets must have their paths setup up by some controller and so there should not be any free flowing packets or messages.
This comment was marked helpful 0 times.
nrchu
@dtaveras: he's talking about the "probing" which, I'm assuming, is essentially done with a packet. You send a packet down to "scout" a path out, but there might be other "probes" that are scouting paths too, and it seems like there could be a lot of possible problems with contention for path building. That's what the slide seems to imply, anyway. If there was a central controller that has a personal "map" or something like that keeping track of all the routes, then it looks like we'd avoid any possible race conditions or deadlock, but you also just introduced a single point of failure in your system which may be bad for other reasons.
Question: How is circuit-switched routing actually implemented? It seems really hard to have probing happen both in parallel and concurrently with other packets coming through the network. Do you associate a lock with each node in the network and just try to take locks along the path? This seems like it would be wrought with inefficiency.
This comment was marked helpful 0 times.
@wcrichto I'm not sure what you mean by packets coming through the network conflicting with probing because if its a circuit switched network all packets must have their paths setup up by some controller and so there should not be any free flowing packets or messages.
This comment was marked helpful 0 times.
@dtaveras: he's talking about the "probing" which, I'm assuming, is essentially done with a packet. You send a packet down to "scout" a path out, but there might be other "probes" that are scouting paths too, and it seems like there could be a lot of possible problems with contention for path building. That's what the slide seems to imply, anyway. If there was a central controller that has a personal "map" or something like that keeping track of all the routes, then it looks like we'd avoid any possible race conditions or deadlock, but you also just introduced a single point of failure in your system which may be bad for other reasons.
This comment was marked helpful 0 times.