Self-Management in Chaotic Wireless Deployments

1
Self-Management in Chaotic Wireless Deployments

• A. Akella, G. Judd, S. Seshan, P. Steenkiste
• Presentation by Zhichun Li

2
Overview

• Chaotic Wireless Networks
• Related Work
• Analysis of performance
• Proposed algorithms
• Conclusion

3
Chaotic Wireless Networks

• Unplanned networks deriving from individual
  deployments
• Unmanaged networks often using the same channel
  and not taking care of power control

Self-Management as automatic configuration of key
access point properties
4
Related work

• Some existing software for network management,
  but designed for large scale networks
• Rate control existing algorithms but not in
  conjunction with power control
• Some algorithms reduce power usage to extend
  battery life
• Chaotic network is different from ad hoc networks
  (limited mobility, sufficient power, competition
  for bandwidth and spectrum)

5
Data sets used

• Place Lab 802.11b APs located in various US
  Cities, allows devices location by using radio
  beacons
• Pittsburgh Wardrive based on a few densely
  populated residential areas, it provides
  Geographic coordinates, ESSID, MAC address,
  Channel Used
• WifiMaps provides Geographic Information Systems
  maps, for each AP it has info about Geo
  coordinates, zip code, ESSID, Channel employed,
  MAC address

6
WifiMaps.com
7
Some observations APs density, channels,
802.11b vs. 802.11g
8
Simulation
GloMoSim Topology
9
Simulation assumptions

• Each node on the map is an AP
• Each AP has D clients with 1 D 3
• Clients are within 1 meter from their AP and they
  dont move
• All APs transmit on channel 6
• All APs use fixed power level of 15dBm
• All APs transmit at fixed rate 2Mbps
• RTC/CTS is turned off (default settings)

10
Simulation runs

• http with thinking time by Poisson distribution
  with mean equal to 5s or 20s
• Comb-ftpi, i clients run FTP transmission
• Results
• 83.3 Kbps average load for Http
• 0.89 Mbps for FTP

11
Stretching the distance D1
Little impact of interference between nodes on
user performance
12
Stretching the distance D3
The performance of both protocols suffers density
13
Stretching the distance increased load
14
Two proposed solutions

• To limit the impact of interference between nodes
  we can
• Use an optimal static allocation of
• non-overlapping channels
• Reduce the transmit power levels

15
Non-overlapping channel assignment

• Using channel 1, 6, 11 from map 2a we move to map
  2b

16
Non-overlapping channel assignment
Three non-overlapping channels
Only channel 6
17
Transmit power control
Transmit power reduced to 3dBm
18
So

• End-user performance can suffer significantly in
  chaotic deployments, especially when there is
  aggressive use of network
• Managing power control and using static
  allocation of non-overlapping channels can reduce
  the impact of interference on performance

19
Problems need to solve

• By reducing the transmission power, we face a
  tradeoff between interference and throughput of
  the channel, since the transmitter is forced to
  use a lower rate to deal with the reduced
  signal-to-noise ratio
• Chaotic networks independent users or
  organizations (often 1 AP) that want to transmit
  always at highest power with suboptimal results
  in terms of performance

20
Ideal solution

• Algorithms socially responsible that act for
  the good of the entire area and reduce their
  power appropriately
• Different from other algorithms that require
  global coordination between multiple APs
• New power control management could be quickly
  spread due to the high rate of deployments of
  802.11g

21
Proposed algorithms

• PARF Power-controlled Auto Rate Fallback
• Based on ARF
• It Attempts to elect the best transmission rate
• If a certain number (6) of consecutive packets
  are sent successfully, the node selects the next
  higher transmission rate
• If a certain number (4) of consecutive packets
  are dropped, the node decrements the transmission
  rate
• Extension of ARF by adding low power states above
  the highest rate state. Power is repeatedly
  reduced until either the lowest level is or the
  transmission failed threshold is reached

22
Proposed algorithms

• PERF Power-controlled Estimated Rate Fallback
• Based on ERF
• It uses path loss information to estimate the SNR
  with which each transmission will be received
• It tries the rate immediately above the estimated
  transmission rate after a consecutive successful
  send
• If the estimated SNR is above a certain amount
  the decision threshold for the highest transmit
  rate, the transmission power is reduced to
  estimatedSNR decisionThreshold powerMargin

23
PERF evaluation
24
Conclusion

• Power control and rate adaptation can reduce
  interference between nodes in a dense wireless
  network
• Implementing those management algorithms in
  commercial APs it is possible and it would spread
  quickly

25
Questions?