Bridging Theory and Practice in Wireless Networks

1. Bridging Theory and Practicein Wireless Networks Theodoros Salonidis

2. Wireless landscape • Technologies • 802.11 (a/b/g/n/s) • 802.16 (e/j/m) • CWUSB (WiMedia UWB) • Bluetooth • Zigbee • Cognitive • Proprietary • Applications • Video, Voice, Data, Internet access, Mobility • Where? • Home • Enterprise • Communities • Sensornets • How? • Single-hop / Multi-hop • Infrastructure / Ad Hoc

3. High-level challenge • Goal: Enable and optimize a variety of emerging wireless applications using a variety of technologies and environments • We need both theory and practice to achieve this goal • However: Currently large gap between the two. • Where should we start from? Practice? Theory? • What are the research problems that need to be addressed to bridge this gap?

4. Practice to Theory? • Some wireless network enablers • Application: BitTorrent • Transport/Congestion control: TCP/IP • Network: link state / distance vector • MAC: ALOHA => Ethernet => 802.11 • PHY: TURBO codes • All were initially invented based on intuition (and often need!). Not theory-driven. • “They work” • but many performance problems when applied in different wireless environments and applications • All have inspired a lot of research • Theory: Models (e.g. TCP, 802.11 MAC) • Systems: Solutions for their problems when applied to different environments

5. Theory to Practice? • Some wireless theory drivers • Asymptotic Capacity Analysis (Gupta-Kumar) • Transport: Network Utility Maximization (Kelly) • Network: Network flow routing (Gallagher) • Link: Backpressure/max-weight scheduling (Tassiulas/Ephremides) • Most based on abstract models. Not yet fully translated to real systems. • We don’t know yet how to build wireless networks that achieve well-defined performance objectives.

6. Theory-driven wireless network design • Derive intuition from theory results. • Implement using heuristics that overcome impractical components but consistent with the theory guidelines. • Result: Improved performance over baseline protocols. • Recent examples: • wGPD (Akyol et. al, 2008) • DiffQ (Warrier et al., 2009) • Horizon (Radunovic et al., 2008)

7. Example: Network Utility Maximization (NUM) xf4 xf1 f1 f4 f3 xf3 f5 xf5 xf2 f2 • Objective • Utilize the full network capacity • Regulate rates xf to operate the network at a well-defined “optimal” performance point at all times

8. NUM formulation Uf(xf) Maximize f - mji(f) xf mij(f) - s.t. for each node i f in Fi f,j f,j mij(f) for each link ij in capacity region L f in Fij

9. NUM Solution • Transport: Source rate control • Network: Flow scheduling • MAC: Back-pressure / Max-weight scheduling • PHY: ?

10. Back-pressure / Max-Weight scheduling Capacity Region L Multi-hop Wireless Network (qi(t)-qj(t)) Cij( I(t), S(t) ) Maximize i,j Topology State Queue Difference Control Action

11. Back-pressure / Max-Weight scheduling Capacity Region L Multi-hop Wireless Network (qi(t)-qj(t)) Cij( I(t), S(t) ) Maximize i,j Topology State Queue Difference Control Action • Several assumptions • Perfect synch + centralized scheduling + knowledge of S(t) + I(t) within capacity region + I(t)??

12. Some open “bridging” research problems • Link quality and interference estimation • Theory says it can make a huge difference in capacity. • What layer should estimations be performed? • “Max-weight” MAC protocols • Possible with today’s hardware platforms but need to be designed very carefully! • “Centralized not scalable”. How about 100 nodes? • Estimation of capacity region for “imperfect” MAC protocols • Three dimensions: estimation accuracy, incorporation to optimization problems, and online operation • Incorporation of additional resources • Multi-antennas (space), multiple channels (frequency) • Delay-aware solutions • Theory still lagging but also plenty of room for new practical solutions that may (or may not) inspire it.

13. Internet Reality example: TDMA MAC mesh architecture • All nodes are slot-synchronized • Periodic frame • Conrol Subframe (CS): Management functions • Data Subframe (DS): Conflict-free data transmissions at each slot MCU GW MAP • Network transmission schedule • Computed by Mesh Control Unit (MCU) • Advantages • Bandwidth and delay guarantees • All MAPs get fair bandwidth share CS DS … … … Frame k

14. TDM protocol design principles • Problem • Determine MAC protocol design parameters subject to hardware and synch bottlenecks • Our contributions • Model: Relates hardware bottlenecks to MAC protocol design parameters through a set of constraints • Measurement methodology to determine the bottlenecks • Synchronization algorithm • Thomson AP implementation • Currently supports 300us slot at 54Mb/s Thomson Max/Eagle Mesh Access Point (MAP) Thomson Max/Eagle MAC Architecture

15. Internet Slot overhead Frame overhead Challenges • Hardware bottleneck analysis • Determine MAC protocol parameters that minimize overhead MCU GW • Clock synchronization • High-accuracy (us), robust, fast MAP • Interference estimation • High-accuracy, fast • Scheduling decision • Optimal, fast CS DS … … … • Schedule dissemination • Robust, fast Frame k

16. Summary • Wide spectrum of research problems • Plethora of wireless technologies and applications • Large momentum in industry/economic sector • Systems wireless networks research just started • Much theory/simulations in the past but open and affordable platforms only recently available • Wireless networking research elements • Experimentation on wireless platforms indispensable • Focus on applications but no need to invent killer-apps • Design application-oriented protocols and solutions that combine existing/emerging wireless link technologies • Put existing theory to the test