Architectures and Algorithms for Future Wireless Local Area Networks

1. Architectures and Algorithms for Future Wireless Local Area Networks PhD Defense 2012-12-14 Peter Dely

2. High speeds require small distances Carrier: 2.4 GHz Modulation: BPSK Channel: 20 MHz 7000 Mbit/s Carrier: 60 GHz Modulation: QAM256 Channel: 1800 MHz 2 Mbit/s 2

3. AP selection with small WLAN “cells” Case 1: AP selection for mobility management 2

4. AP selection with small WLAN “cells” Case 2: AP selection for load balancing 2

5. Research questions • Which AP should a station select at a given time? • How to design and implement an architecture that enables fair resource distribution and mobility management? • How to estimate the quality of an AP?

6. Greedy is not always best Handover Handover 6 Mbit/s 54 Mbit/s Given a certain “cost” for a handover, when is it beneficial to perform that handover? 4

7. Dynamic model finds best AP for the known future Important parameters 1. Cost of a handover 2. Link rates between APs and stations 3. Interference Important constraints 1. Capacity of the wireless channel 2. Data transfer only after a station has been connected to an AP for a certain time Describe system characteristics Describe system state Mixed Integer Linear ProgramSolver Maximize minimum average throughput Objective Compute 1. Which station uses which AP at what point in time 2. Which rate it can use for downloading 4

8. Evaluation scenario • Comparison of • Greedy • Hysteresis • k-Handover • Mobile stations “walk” on the corridors • Random waypoint with fixed waypoints • Comparison metric: • Normalized throughput • Relative to optimal solution with dynamic model • 1 = Optimum 4

9. Impact of mobility Impact of handover cost Mobility and handover costs reduce performance More mobility = Less performance Higher handover costs ~ Less performance 4

10. AP selection in wireless mesh networks Internet 5

11. Max-min fair rate allocations and single path routing cannot be found through a simple MILP Important parameters 1. Network graph 2. Interference set 3. Association opportunities Important constraints 1. As in the pure AP scenario 2. Flow conservation constraints 3. Single-path routing constraints Max-min fair optimization is multi-objective optimization Standard algorithms are very slow  Design of heuristic MESHMAX-FAST(*) Describe system characteristics Describe system state ReformulateModel Mixed Integer Linear ProgramSolver Solution algorithm Solve partial problem Objective Max-min fair rate allocation Compute 1. Which traffic is sent via which route 2. Rates between the stations and the Internet 3. Which station uses which AP 5

12. MESHMAX-FAST* is better than a straight forward linear relaxation In 50% of cases, the performance is greater than 95% of the optimum Better 5

13. MESHMAX-FAST is suitable for online optimization Optimal solution: > 1000 seconds MESHMAX-FAST*: < 10 seconds with comparable results 5

14. Research questions • Which AP should a station select at a given time? • How to design and implement an architecture that enables fair resource distribution and mobility management? • How to estimate the quality of an AP? 6 7

15. A software defined mesh network Internet Monitoring and Control Server OpenFlow Control Server • Build network graph • Run MESHMAX-FAST • Trigger actions at the OpenFlow control server • Programmable forwarding unit (OpenFlow) • Legacy routing protocol in a virtual network • Configure routes • Configure rate shapers • Trigger handovers 6

16. Testbed setup IEEE 802.11a links, fixed PHY Download from “Internet” Gateways are connected via “DSL” Results with 6 Mbit/s GW links SNR = Use AP with best signal Hop-count = Use AP closest to GW MESHMAX = Use optimal AP MESHMAX performance in real networks 10 4 5 1 2 3 6 8 9 7 6

17. CloudMAC distributes MAC processing Virtual AP Virtual AP Virtual AP OpenFlow Controller Virtual WLAN NIC Virtual WLAN NIC Virtual WLAN NIC Internet Application Application Policy Program switch 7

18. CloudMAC enables simple handovers Virtual AP Virtual AP Virtual AP OpenFlow Controller Virtual WLAN NIC Virtual WLAN NIC Virtual WLAN NIC Internet Application Application Policy Program switch Change flow table 7

19. CloudMAC reduces data loss during handovers Median reduction in packet loss from approx. 10000 to 3.5 • Standard IEEE 802.11: scan and re-association • CloudMAC: Association state in Virtual AP  no reassociation 7

20. Research questions • Which AP should a station select at a given time? • How to design and implement an architecture that enables fair resource distribution and mobility management? • How to estimate the quality of an AP? 8 9

21. Finding the AP with the best link is hard Interference Cross traffic 8

22. RSSI is not a good indicator for throughput 100% loss 0% loss 8

23. BEST-AP uses regular traffic for performance estimation Internet BW Estimation Handover BW Estimation BEST-AP Server 8

24. Testbed setup Two APs Interference according to real traces Dynamic AP selection with stationary users Results Interference Interference 8

25. Evaluation with mobile users Video freeze duration Quality for current AP starts to decrease Video playout resumes Buffer Level Video playout freezes Connect to new AP Connection completely breaks Scan for new APs Time 9

26. BEST-AP reduces video freezes Testbed setup Results • BEST-AP has fewer and shorter freezes than Linux • A dedicated scan card is not necessary 9

27. Summary of contributions • Theoretical study of AP selection problem • Considering handover costs • Wireless mesh backhauls • Architecture proposals and their implementation • For the centralized management of wireless mesh networks • For distributing MAC layer processing • Quality estimation of APs • Bandwidth estimation method • Improved video streaming 10

28. Future directions • Use other modeling frameworks • Convex optimization • Robust optimization • Unify architectures • Use CloudMAC in mesh networks • Integrate bandwidth estimation in system • Comprehensive tests in large networks • More realistic traffic loads • Scalability tests 10

29. Thank you! Watch this talk on youtube.com/kaumesh

30. Backup slides

31. … but this comes at a cost 36 40 44 48 52 56 60 64 2.4 GHz • From 2.4 GHz to 60 GHz  Higher path loss • Higher attenuation by walls etc. • From BSPK to QAM256 • More constellations • More susceptible to noise • From 20 MHz to 160 MHz channels  Less spatial reuse  Higher energy use 60 GHz To achieve high speeds short distances between the AP and the stations are necessary. BSPK QAM256 2

32. Using predictions to improve performance • Main idea: use predictions of the future for optimization • Optimize each time slot using the estimate of the future More estimationerrors 4

33. MESHMAX-FAST improves performance 5

34. Key ideas of MESHMAX-FAST • Observation: APs are traffic aggregation points for many stations • Algorithm sketch: • Compute a multi-path solution between the APs and the stations • Assign stations to APs according to the capacity of the APs • Re-compute the multi-path solution and give priority to APs with many stations • Move stations to other APs if it increases the minimum rate • Repeat steps 3-4 until the minimum rate cannot be increased anymore • Finally, find a single-path for each station 5