Traffic splitting algorithms in multipath networks: Is the present practice good enough?

This paper examines different traffic splitting strategies for multipath IP networks. The question is whether today's widely deployed algorithms perform well in all situations. We have implemented in a simulator and analyzed different well-known algorithms. Throughout our study several kinds of traffic traces have been used: real ones captured at backbones as well as at access networks, and specially derived, hypothetical traces. Based on the results we argue that the answer to the leading question is fairly negative: there can be several realistic scenarios, where more sophisticated methods seem to be required.