Wednesday, August 23, 2017

Keynote: Hitting the Nail on the Head: Interdisciplinary Research in Computer Networking -- Prof. Jennifer Rexford

Jennifer Rexford
2016 ACM Athena Lecturer Award Winner

  • The internet is one of the most influential inventions.
  • However, we cannot do it alone.
  • Hammers (techniques) and Nails (problems).

  • Project 1: Protocols as Distributed Optimizers
    • Traditional traffic management has architectural limitations -- slow adaption
    • TCP
      • Forward engineering of traffic management
      • Decompose to generate a distributed solution
    • Lesson: 
      • Start with a well-stated problem, then decompose into a distributed solution
      • Research as decomposition
      • People and timing

  • Project 2: Composition of Network Policies
    • Simple, open data-plane API -- openflow
    • Moduler controller applications -- each module partially specifies the handling of the traffic
    • Abstract openflow -- policy as a function
    • Lesson: 
      • Abstraction and decomposition; 
      • Embedded people (and code)
  • Project3: Traffic monitoring in the data plane
    • Traffic analysis in the data plane
      • Streaming algorithms
      • A great opportunity
    • Example: heavy-hitter detection
      • Approximating and approximation
    • Lesson: 
      • Getting concrete
        • Computational model
        • Example problem
        • Strawman solution
      • Iteratively designing
      • Striving for general understanding
Should we all do interdisciplinary research?
  • Interdisciplinary fun
    • Bigger impact & social fun
  • Managing risks
    • One vs many hammers
    • Steep learning curve
      • Learn both the hammer and the nail
      • Learning the culture of both research fields
    • Join an emerging community
    • Credit for the work
    • Healthy interdisciplinary collaborations
Q: We have various network devices, what is the network constrained by now? What is the next problem in the network?
A: There are still a lot of things to understand. For example, which things should be where.