Network Management

1. introduction motivation major components wired network management Internet management infrastructure tomography (using end-to-end measurement) wireless network management enterprise WLAN chaotic WLAN Network Management

2. Self-Managementin Chaotic Wireless Deployments Authors: Aditya Akella, Glenn Judd, Srinivasan Seshan, Peter Steenkiste MobiCom 2005

3. introduction characterizing current 802.11 deployments impact on end-user performance limiting the impact of interference power and rate selection algorithms conclusions Outline

4. Introduction • chaotic deployment • unplanned • highly variable AP densities • unmanaged • not configured to minimize interference • questions: • impact of interference on end-user performance? • how to improve end-user performance in chaotic deployments?

5. Related work • management in wired networks [IETF zeroconf, Thomson 1998, Droms 1997] • rate adaptation in ad-hoc networks [Sadeghi 2002] • traffic scheduling in sensor networks and 802.11 networks [Qiao 2003, Kompella 2003] • power and rate control in ad-hoc routing protocols [Kawadia2005, Draves 2004, Santhanam 2003, Holland 2003] • commercial products

6. introduction characterizing current 802.11 deployments impact on end-user performance limiting the impact of interference power and rate selection algorithms conclusions Outline

7. Characterizing current 802.11 deployments • measurement data sets • Place Lab • WifiMaps • Pittsburgh Wardrive • measurement observations • 802.11 deployment density • 802.11 channel usage • 802.11b vs. 802.11g • vendors and AP management support

8. 802.11 deployment density Data set: Place Lab Degree: # of neighbor APs (within 50m) Deployment: high density Degree≥3 (interference)

9. Measurement observations • 802.11 channel usage • 802.11b vs. 802.11g • 20% are 802.11g • vendors and AP management support Channel usage Popular AP vendors

10. introduction characterizing current 802.11 deployments impact on end-user performance limiting the impact of interference power and rate selection algorithms conclusions Outline

11. Simulation topology • D clients with an AP • Clients 1m away from AP • APs on channel 6 • transmit power: 15 dBm • transmission rate: 2Mbps • RTS/CTS turned off • two-ray path loss model • Ricean fading model Data set: Pittsburgh Wardrive

12. Simulation set-up • HTTP • client run HTTP with AP • two HTTP transfers separated by a think time (Poisson distribution) • comb-ftpi • i clients run long-lived FTP

13. Interference at low & high client densities • interference increases with client density • more degradation when traffic load is high One client per AP Three clients per AP

14. introduction characterizing current 802.11 deployments impact on end-user performance limiting the impact of interference power and rate selection algorithms conclusions Outline

15. Limiting the impact of interference • optimal static channel allocation • transmit power control

16. Optimal static channel allocation • optimal channel allocation helpful, but cannot eliminate interference single channel three channels

17. Transmit power control power level: 15dBm power level: 3dBm optimal channel allocation + transmit power control optimal channel allocation • transmit power control improve application performance, and network capacity & fairness

18. introduction characterizing current 802.11 deployments impact on end-user performance limiting the impact of interference power and rate selection algorithms conclusions Outline

19. Power and rate selection algorithms • benefits of transmit power reduction • fixed-power rate selection algorithms • Auto Rate Fallback (ARF) • Estimated Rate Fallback (ERF) • power-controlled rate selection algorithms • power-controlled Auto Rate Fallback (PARF) • power-controlled Estimated Rate Fallback (PERF) • performance evaluation

20. Benefits of transmit power reduction distance between client and AP: 10m • lower transmit power supports higher AP density • determine transmit power for a given AP density control to achieve a certain throughput

21. Fixed-power algorithm 1: Auto Rate Fallback (ARF) • intuition: a failed transmission indicates transmission rate too high • a number of packets transmitted successfully => select higher transmission rate • a number of packets dropped => decrease transmission rate • idle for a certain amount of time => use the highest possible transmission rate for next transmission

22. Fixed-power algorithm 2: Estimated Rate Fallback (ERF) • determines highest transmission rate based on SNR • estimate SNR: tag transmission power, path loss and noise estimate in packets • SNR = txPower – pathloss – noise • accommodate uncertainty in SNR measurements

23. Power-controlled algorithms • each AP acts socially • reduce transmit power (interference to other APs) as long as not reduce its transmission rate • power-controlled Auto Rate Fallback (PARF) • at certain rate, reduce power level after a number of successful sends • power-controlled Estimated Rate Fallback (PERF) • reduce transmit power while maintain the required SNR for the transmission rate

24. Performance evaluation • effect of power & rate selection algorithms used by aggressor pair on victim pair

25. Performance evaluation aggressor-pair rate unlimited aggressor-pair rate limited • PERF almost eliminates the interference on the victim pair

26. Conclusions • chaotic networks • unplanned • unmanaged • reduce interference while ensuring robust end-client performance • PERF: reduces transmission power as much as possible without reducing transmission rate

27. Wireless network management: summary • Reading list • enterprise WLAN management: a drastically different approach • sensor network management • Future research • Management architecture? • Tomography-based approach?