Bounds on end-to-end performance via greedy, multi-path routing in integrated services networks

We focus on the design and performance of resource management controls, specifically state dependent routing, with which to compute the bounds on the end-to-end performance in integrated services networks. The routing control problem at each hop is transformed into an optimization problem, the solution of which suggests a path selection algorithm that relies on the available capacity process at adjacent nodes. We focus on computing a bound on the average end to end packet loss, although our results can easily be adapted to minimize the mean transfer delay. The routing scheme based on the solution to the optimization problem is as follows: each node computes multiple alternate paths to a given destination. The available capacity process, e.g., available bandwidth, available buffer, etc., is derived from the link state updates received at each node and the forwarding rule is based on some function of the maximum available capacity process applied to the set of possible nodes corresponding to the next hop on an alternate path for a particular source destination pair. A consequence of this path selection rule is the pooling of network resources when the load is increased. This allows us to prove the bound (at each hop) on the packet loss in terms of the network parameters. We show how the node reduction technique (and its associated bound) can be used as a building block to compute end to-end performance bounds in a network setting.