QoS routing relies on both the knowledge of the connection's requirements and the information about the availability of resources in the network. Due to interconnections, imperfect measurements, aggregation in large networks and the statistical nature of the resources costs, inaccuracies in the information available for making routing decisions are unavoidable. Hence, it is possible that during the connection will be rolled back to route the call via an alternate path. Crankbacks that are caused by link or node failure or by insufficient non-additive metrics (e.g., bandwidth), can be predicted by checking the path using the local view of intermediate nodes, and discerning that not all the nodes along the rest of the path can support the required QoS. However, predicting a crankback caused by additive metrics (e.g., delay), which is the focus of our work, is more complicated. This requires nodes to identify an additive metric resource that is running out too fast, and perform a crankback before the value of an end-to-end QoS requirement is violated.

In our presentation, we propose a novel technique to improve the crankback mechanism in modern communication networks such as MPLS. Under the proposed scheme, during the connection setup phase, each node along the selected path evaluates the expected time to successfully establish a connection that satisfies all the QoS requirements. If the current path does not minimize the expected setup time, crankback is initiated even before a certain QoS parameter is violated.