Duration: 1:13:31

38 views

0

The Internet's TCP/IP stack was not designed with the vagaries of the wireless channel and mobility in mind and is showing signs of age today. TCP has well-known problems over multi-hop wireless networks as it con?ates congestion and loss, performs poorly over time-varying and lossy links, and is fragile in the presence of route changes and disconnections. In this talk, I will present a clean-slate design and implementation of a wireless transport protocol, Hop, that uses reliable per-hop block transfer as a building block. Hop is 1) fast as it eliminates many sources of overhead as well as noisy end-to-end rate control, 2) robust to partitions and route changes because of hop-by-hop control as well as in-network caching, and 3) simple as it obviates complex end-to-end rate control as well as complex interactions across layers. Our experiments over a 20-node multi-hop mesh network show that Hop is dramatically more efficient, achieving better fairness, throughput, delay, and robustness to partitions over several alternate protocols, including gains of over an order of magnitude in median throughput. I will present arguments for why block transfer forms the basis of a simple, robust protocol stack that gracefully degrades in performance across diverse wireless environments such as access point networks, meshes, mobile ad hoc networks, and DTNs. I will outline research challenges and opportunities raised by block transfer in order to realize this vision.