"Promising approaches to clean-slate wireless networking" by Chris Ramming

April 24, 2008, Thursday, 12:15-1:30pm:
Attended this Lunch time talk hosted in Stanford Packard 101. I was lost in the east side of the campus and walked in the room a few minutes late.

The topic is quite interesting. It looks at the MANET from a DARPA program manager's view and summarized the work related to the MARCONI program. Majority of the MANET are concentrated on cross-layer design.

At a higher level, the whole network design problem are modeled as an optimization problem with an objective function subjects to a set of constraints. I like this type of high-level view. It captures both the primal and dual sides of the problem. Eventually, narrows down by using the elastic and inelastic traffic to decompose the objective function into two sub-objectives. Inelastic sub-objective requires a minimum rate guarantee plus a latency bound. It was summarized that the DoD network contains more than 80% of these inelastic traffic which should be paid for more attention to.

Speaker also mentioned some of the research work on network coding. One
interesting conclusion is that network coding enables the traffic flow through the weakest cut-set of the graph, instead of the weakest link comparing to packet switched network. To my understanding, network coding is just another approach to implement maximum flow over the network in much smaller granularity. The benefit of maximum throughput will bring problems on the scalability to manage large number of flows over the shared network. I asked question regarding the difference between P2P and network coding. Chris said there is one P2P implementation using network coding idea. From the optimized load balancing or throughput maximization's point of view, they are just different implementation approach to me. However, all the nice ideas are welcome in the research domain.


This talk is given in an excellent format at the right technical level for me. Good I was there.

The original abstract is below:
====
About the talk:

Mobile ad-hoc networks (MANETs) offer a particularly thorny challenge for network architects and protocol designers. To first order, radio phenomena are a poor fit for traditional link-and-node network abstractions not only because of broadcast but also because of the potential for constructive and destructive interactions between nodes at a distance. Moreover, MANET protocol designers must address a regime that starkly contrasts with experience in the wired Internet insofar as spectrum is scarce, the medium is relatively prone to error and erasure, and end-to-end latency tends to be relatively high. And finally, network management must be intrinsic to the protocol suite rather than a bolt-on afterthought, because MANETs are envisioned for use in dynamic settings that do not admit of manual operating processes. It should be no surprise that protocols and architectures designed for static, reliable, high-bandwidth settings are poorly suited for use in MANETs; the question is what alternatives should be pursued and in what way. Are there practical systematic strategies for clean-slate network design? This talk will frame MANETs as a challenge in clean-slate design, and will describe how some recent experimental projects have made demonstrable progress and gained experience in first-principles network design by applying concepts and strategies such as network coding and optimization decomposition.