I understand. Routing policy can result in TIVs, if the policy considers factors other than latency. I think the technical limitation could be the reason of TIVs too. Resolving TIVs caused by technical reasons in IP layer may be hard, but resolving with relay is easier.
Take AS routing as an example. Suppose both AS B and AS C can route traffic from AS A to AS D. AS B charges a higher price for routing traffic, but routing through AS B is fast. Suppose AS A's operator's will is: a flow to AS D should route through AS C (slow but cheap) if QOS meets the application's requirement, otherwise route through AS B (fast but expensive). If using routing policy only, out-bound router must sense the exact QOS requirement of every traffic flow and the latency of AS path. So only configuring routing policy at IP layer may not fulfill the will of operator entirely. Both routing policy and relay should be used to fulfill operator's will. On Tue, Oct 27, 2009 at 7:23 PM, Enrico Marocco <[email protected]> wrote: > lichun li wrote: >> Indeed, TIVs should be addressed in routing layer. However, solving >> the problem in routing layer may involve things like changing routing >> protocol and replacing routers, which will pay some time and high >> price. In contrary, relay is an easier solution for applications >> sensitive to delay, and is already used somewhere. > > I'm sorry you lost me here. TIVs are the result of routing policies, I > don't think there's anything wrong with current routing protocols, at > least not in a way that significantly influences TIVs. > >> Relay can cause layer violation. Operators may not like it. But >> operator's participating in relay selection may reduce the harm of >> layer violation. > > If a routing policy makes traffic flow from AS A to AS B on a slow yet > capable direct link instead of rerouting it through a fast connected AS > C, well, I'm pretty sure that policy entirely reflects the operator's > will. If they were actually interested in making traffic follow > different paths, they'd just need to reconfigure the routers without > recurring to application level relay (that, BTW, basically doubles > bandwidth costs) and relay selection. > > -- > Ciao, > Enrico > -- Sincerely yours, Lichun Li Mobile Life New Media Lab, BUPT _______________________________________________ alto mailing list [email protected] https://www.ietf.org/mailman/listinfo/alto
