Opportunistic Use of Client Repeaters to Improve Performance of WLANs

1
Opportunistic Use of Client Repeaters to Improve
Performance of WLANs
Victor Bahl1, Ranveer Chandra1, Patrick P. C.
Lee2, Vishal Misra2,Jitendra Padhye1, Dan
Rubenstein2, Yan Yu31Microsoft Research2Dept
of Computer Science, Columbia University3Google
Inc.Dec 12, 2008
2
Outline

• Rate anomaly problem
• SoftRepeater design
• Fairness requirements
• Experimental results
• Conclusions

3
Rate Anomaly of 802.11
54Mbps
18Mbps

• Rate anomaly is well-known in WiFi 802.11
  networks
• Low-rate stations degrade throughput of high-rate
  stations
• Why does rate anomaly exist?
• Stations reduce data rates when signal strength
  is poor (auto-rate)
• Low-rate stations packets consume more airtime
• 802.11 arbitrates transmissions on per-packet
  basis
• High-rate stations receive limited airtime ?
  throughput degrades

54Mbps
A far from AP
A, B near AP
4
Limitations of Prior Solutions

• Whats new? Rate anomaly is well-known, with many
  solutions proposed.
• Assumptions of prior solutions
• Require dedicated hardware (e.g., Cisco Aironet
  1200 series APs)
• Change MAC layer (e.g., Lee et al., Infocom 04
  Liu et al., JSAC 05)
• Construct ad-hoc mesh networks (e.g., Draves et
  al., Mobicom 04)
• Drawbacks of prior solutions
• More cost for hardware change
• Not compatible with widely deployed
  infrastructure networks
• Inflexible solutions cannot be activated on
  demand

5
Our Solution SoftRepeater

• SoftRepeater A practical, deployable system that
  addresses rate anomaly
• Main idea
• High-rate station (repeater) relays traffic for
  low-rate station (client)
• Key features
• Repeater is opportunistic - activated only when
  both repeater and client receive beneficial
  throughput
• No changes to 802.11 MAC and AP
• Deployable in infrastructure and adhoc networks

6
Design Issues

• How can we detect existence of rate anomaly
  occurring?
• How do we formally define beneficial
  throughput?
• How do we support multiple interfaces on a
  wireless card?
• We need managed mode for communication between AP
  and repeater
• We need adhoc mode for communication between
  repeater and client
• What fractions of time should we give to
  managed/adhoc modes to ensure beneficial
  throughput?

7
Our Contribution

• Propose a handshaking protocol for detecting rate
  anomaly and reaching consensus on using
  SoftRepeater
• Formalize a set of utility maximization problems
  for different fairness requirements
• Implement SoftRepeater on Windows XP conduct
  extensive testbed experiments and QualNet
  simulations

8
SoftRepeater Architecture

• Built on VirtualWifi allowing two virtual
  interfaces for a wireless card
• Primary Virtual Interface communication between
  AP and repeater in managed mode
• Repeater Virtual Interface communication
  between repeater and client in adhoc mode
• Repeater Virtual Interface activated only when
  beneficial to both repeater and client
• Alternate between primary and repeater interfaces
  with switching overhead lt 40ms
• Optional Network Coding Engine that further
  boosts throughput, with slight modifications to
  AP
• Multiple radios can be supported (not in our
  current experiments)

9
Detecting Rate Anomaly

• Goal Determine When SoftRepeater is beneficial
• Key steps
• Collect information from nearby stations in
  promiscuous mode
• Number of packets transmitted
• Average size of packets
• RSSI
• Data transmission rate
• BSSID
• Check utilization of medium. If neighbors send
  about the same number of packets, but at a low
  rate, rate anomaly may exist.

10
Repeater Utility Function

• Goal capture throughput gain of both repeater
  and client
• Define a fraction of time spent in managed mode
• Assumptions
• Stations always have backlogged data to send
  (i.e., saturated case)
• Implying equal channel access
• Good approximation for file-transfer applications
• Zero switching overhead
• 1 - a fraction of time spent in adhoc mode
• Can easily account for non-zero switching
  overhead
• Intuition if utility improved for both repeater
  and client, activate SoftRepeater

client
repeater
11
Repeater Utility Function
TA

• Without SoftRepeater
• Bs throughput TB
• As throughput TA
• With SoftRepeater
• Bs Throughput aTB/ 2
• As throughput min(aTB/ 2 , (1- a)TA,B)
• TA,B inferred throughput between A and B from
  RSSI measurement
• If max-min fairness is used, repeater utility
  function becomes
• T maxa minaTB/ 2, min(aTB/ 2 , (1- a)TA,B)
• If T gt TA and T gt TB (better for both) activate
  SoftRepeater

client
TA,B
TB
B
repeater
12
Generalizing Repeater Utility Function

• For different objectives
• Maximizing total throughput starve client (bad)
• Max-min fairness
• Proportional fairness
• For different settings
• In presence of interfering nodes
• In presence of multiple clients
• Multiple radios
• Multiple wireless cards
• Details in paper and tech report

13
Repeater Initiation Protocol

• Goal confirm and reach consensus on activating
  SoftRepeater
• For now simple 4-way handshake
• B broadcasts SoftRepeater offer
• A infers data rate from A to B (from RSSI) and
  unicasts response
• B picks clients to serve (if utility improved)
  and broadcasts final Take it or leave it offer
• A unicasts accept/reject

2. unicast response
4. unicast accept/reject
client
1. broadcast offer
B
3. broadcast new offer
repeater
14
Testbed Experiments

• SoftRepeater is implemented on Windows XP
• Testbed experiments in office building

• AP located at X
• Repeater (node R) fixed at Y
• Client (node C) moved between Y, T, Z

• Use 802.11a, with auto-rate feature enabled
• Focus on Max-Min fairness

15
Experiment 1 Downlink UDP
rate anomaly scenario
AP
C
R

• UDP throughput improved by 200 with SoftRepeater
  when rate anomaly exists

16
Experiment 2 Downlink TCP
rate anomaly scenario
AP
C
R

• TCP throughput improved by 50 with SoftRepeater
  when rate anomaly exists, even communication
  alternates between managed and adhoc modes

17
Experiment 3 UDP with 2 clients
rate anomaly scenario
AP
R
C1 C2

• UDP throughput improved with SoftRepeater when
  two clients served

18
Qualnet Simulation Effectiveness of Repeater
Initiation Protocol

• AP in office 0
• Client in office 9
• Downlink UDP for both repeater and client

Repeater

• SoftRepeater activated only when there is
  throughput gain

19
Qualnet Simulation Multiple Clients

• AP in office 0
• Repeater in office 3
• N clients in office 9
• Downlink UDP

Repeater

• SoftRepeater improves the baseline throughput by
  more than 65.

20
Summary of Experimental Results

• Main observation throughput significantly
  improved for UDP/TCP flows when rate anomaly
  exists
• More experiments in paper/tech. report
• Correctness of repeater initiation protocol
• Extension with network coding
• Various traffic scenarios
• Qualnet simulation for more complicated
  scenarios (e.g., interfering nodes, multiple
  repeaters/clients)

21
Conclusions

• Propose SoftRepeater, a practical, deployable
  system that addresses rate anomaly problem
• Formulate different utility maximization problems
  for SoftRepeater
• Implement a prototype that demonstrates the
  improvement of SoftRepeater

22
Questions pclee_at_cs.columbia.edu
23
Security Issues

• Security concerns
• Privacy
• End-to-end encryption (e.g., IPsec) can be used
• Greedy/malicious repeaters
• Client monitors channel quits if performance
  becomes worse after SoftRepeater is used
• Conclusion Security is no worse than
  SoftRepeater-free networks