Monday, October 27, 2014

Paper 6: "PRAN: Programmable Radio Access Networks" (Wu, Panda, Li, Shenker)


Presenter: Wenfei Wu

Authors: Wenfei Wu (UW-Madison), Aurojit Panda (UC Berkeley), Li Erran Li (Bell Labs), Scott Shenker (UC Berkeley and ICSI)
Paper: http://conferences.sigcomm.org/hotnets/2014/papers/hotnets-XIII-final26.pdf

Existing Radio Access Network (RAN) architectures are struggling to keep up with the exponential mobile traffic growth. These problems can be traced back to: i) insufficient radio resource coordination, ii) inefficient infrastructure utilization, and iii) inflexible data paths.

To  improve radio resource coordination, Wenfei described a new model to handle cell interference---Wenfei independent scheduling---based on optimization techniques.

Wenfei then described a way to share radio resources by moving base station processing tasks to servers. For bursty traffic, Wenfei explained that they can reserve a bunch of resource preemptively.

Finally, Wenfei described a way to bring flexibility in RAN by facilitating HW/SW programmability (current HW/SW radio are hard to program). For that, he proposed a "SDNish" approach in which the data-plane logic is decoupled from the control-plane logic.

The authors evaluated their framework on a real data set, collected over 3300 base stations for 1 hour. Among others, they profiled resource usage in their servers and show that their optimization is efficient. As an illustration, instead of the theoretical 6000 cores required to process their workload, they only needed 200 cores.
       
Future work:

  • Evaluate joint-scheduling
  • Provide performance guarantee 
  • Develop the programmable data-path


Q: Does PRAN run across providers?
A: It can run accross providers. In this case, each providers should reserve frequency and power so that the signals can be sliced.

Q: Is your control-plane really decoupled from the data-plane?
A: Yes, but there still is a communication between the two planes.