On the benefit of information centric networks for traffic engineering

Current Internet performs traffic engineering (TE) by estimating traffic matrices on a regular schedule, and allocating flows based upon weights computed from these matrices. This means the allocation is based upon a guess of the traffic in the network based on its history. Information-Centric Networks on the other hand provide a finer-grained description of the traffic: a content between a client and a server is uniquely identified by its name, and the network can therefore learn the size of different content items, and perform traffic engineering and resource allocation accordingly. We claim that Information-Centric Networks can therefore provide a better handle to perform traffic engineering, resulting in significant performance gain. We present a mechanism to perform such resource allocation. We see that our traffic engineering method only requires knowledge of the flow size (which, in ICN, can be learned from previous data transfers) and outperforms a min-MLU allocation in terms of response time. We also see that our method identifies the traffic allocation patterns similar to that of min-MLU without having access to the traffic matrix ahead of time. We show a very significant gain in response time where min-MLU is almost 50% slower than our ICN-based TE method.