Multihop Networks: Fact or Fiction? Ramki Gummadi (MIT)
Multihop: Brief history Multihop wireless successful since 60s More recently, cellular repeaters Large-scale mesh more recent CW 2008
Applications Large-scale urban n/w (Meraki, Roofnet) Overcomes last-mile challenges Rural long-distance (WiLDNet) Low-cost, high throughput Sensor n/w (several deployments) Low-power radios Ad-hoc networks Disaster recovery, military, neighborhood mesh CW 2008
Three mesh n/w challenges Theoretical limits Practical considerations Business models CW 2008 Page 4
Theoretical limits 1 p n Mesh throughput degrades as O( ) Intuition: direct communication throughput between s and d goes to 0. So, have to multihop But multihopping increases load at each node 1 p n Rmax Rmin ... s d CW 2008
Workarounds Local communication, gateway nodes Variable-width channels Reduces aggregate load Variable-width channels Reduces interference, exploits spatial diversity Fixed throughput between any two nodes Better radios, algorithms Interference alignment (Grokop,Tse,Yates’08) Interference cancellation CW 2008
Practical considerations Multihop interference more complex than single-hop interference (Rangwala et al.,’06) Current MACs and TCP don’t work well CSMA back-off, carrier sense failure TDMA, interference cancellation TCP starvation, unfairness (Rangwala et al.,’08) Rate-based TCP, network coding 4 X 2 3 X 1 CW 2008
Theory and practice intersection Do multiple radios help? Only by a constant factor But what about imposing structure? Mesh with a tree backbone scales better Do multiple channels help? Again, only by a constant factor But what if we exploit bursty traffic? Better scalability if all channels usable by all nodes CW 2008
Space: The third dimension Poor spatial reuse observed in Roofnet Contributes to low throughput (O(100) Kbps) Omni-directional antennas Directional antennas New development: phased arrays But, reduces the rich scattering MIMO needs PHY and MAC issues less well-understood CW 2008
Business issues Different business models Traditional ISP model: Rice TFA Commodity hardware, modified firmware (Firetide, Tropos, Belair, etc.) Ad-based free-ISP model: Meraki Grassroots deployment CW 2008
Conclusions Rich set of cross-cutting issues Both theoretical and practical Important question: external interference from unmanaged networks Scope for cute algorithms (two out of last four best SIGCOMM papers) Potential to realize the wireless dream Always on, faster, cheaper, better Are we there yet? No CW 2008