> When I say this link will have this bandwidth it sounds to me like it's a > dedicated bandwidth that limits the link to the given value. > When I say priority I think of a link with several clients where one of them > gets priority over the others, but in case there are no others it can get > more.
it's not about clients, but queues. technically there is no problem putting one client in the priority queue and others in normal queues, but that's not the intention behind qos. it's all about applications with different needs (bandwidth, jitter, drop probabilities etc). a longer explanation than olivers is: a class with priority will have a strict priority queue, i.e. traffic in this queue will be serviced before all other queues. this also means that a priority queue can starve other queues, so it's important to limit the traffic in the queue. that is done by a policer[1] which sets a *maximum* bandwidth for the queue. a class with a bandwidth statement is given a *minimum* guarantee for the traffic in the queue. if one or more queues don't use the allocated bandwidth, the excess bandwidth is shared among the other "bandwidth" queues, either in the same proportion as the configured bandwidth or the configured "bandwidth remaining". in essence: * priority <bw> gives a strict priority queue with a *maximum* bw guarantee * bandwidth <bw> gives a queue with a *minimum* bw guarantee [1] this can either be configured as: class X priority <bandwidth> or class X priority police rate <bandwidth> -- Pelle _______________________________________________ cisco-nsp mailing list cisco-nsp@puck.nether.net https://puck.nether.net/mailman/listinfo/cisco-nsp archive at http://puck.nether.net/pipermail/cisco-nsp/
