Challenged Networking : An Experimental Study of new Protocols and Architectures

With the growth of Internet, the underlying protocols are increasingly challenged by new technologies and applications. The original Internet protocols were, however, not designed for wireless communication, mobility, long disconnection times, and varying bandwidths. In this thesis, we study challenged networking, and how well old and new protocols operate under such constraints. Our study is experimental. We build network testbeds and measure the performance of alternative protocols and architectures. We develop novel methodologies for repeatable experiments that combine emulations, simulations and real world experiments. Based on our results we suggest modifications to existing protocols, and we also develop a new network architecture that matches the constraints of a challenged network, in our case, an opportunistic network. One of our most important contributions is an Ad hoc Protocol Evaluation (APE) testbed. It has been successfully used worldwide. The key to its success is that it significantly lowers the barrier to repeatable experiments involving wireless and mobile computing devices. Using APE, we present side-by-side performance comparisons of IETF MANET routing protocols. A somewhat surprising result is that some ad hoc routing protocols perform a factor 10 worse in the testbed than predicted by a common simulation tool (ns-2). We find that this discrepancy is mainly related to the protocols’ sensing abilities, e.g., how accurately they can infer their neighborhood in a real radio environment. We propose and implement improvements to these protocols based on the results. Our novel network architecture Haggle is another important contribution. It is based on content addressing and searching. Mobile devices in opportunistic networks exchange content whenever they detect each other. We suggest that the exchange should be based on interests and searches, rather than on destination names and addresses. We argue that content binding should be done late in challenged networks, something which our search approach supports well