End-to-end performance guarantees for multipath flows

When routing data across a network from one source to one destination, instead of following a fixed path, one can choose to spread data on several routes in order to use all potential ressources of the network. This issue has been studied for many models of networks with various objectives to optimize. In this paper we investigate how to route a flow across a network of servers with end-to-end performance guarantees in the framework of Network Calculus. We discuss stability issues (i.e. whether we can ensure that end-to-end delays are bounded) for arbitrary networks, and how to compute bounds on worst-case end-to-end delays and backlogs. The tightness issues are discussed on a small but challenging toy example.