Methods for restoring MAC layer fairness in IEEE 802.11 networks with physical layer capture

Sachin Ganu, Kishore Ramachandran, Marco Gruteser, Ivan Seskar, Jing DengREALMAN@MobiHoc 2006 · DOI

TL;DR

This paper investigates the "physical layer capture" effect in Wi-Fi networks, where a stronger signal can override a weaker one during collisions, leading to unfair bandwidth distribution. The authors experimentally demonstrate this unfairness and evaluate several methods, finding that adjusting 802.11e EDCF parameters like TxOp and AIFS offers the most promising way to restore fairness.

What Problem Does This Paper Address?

In wireless networks, especially multi-hop ad-hoc networks, ensuring that all users get a fair share of the available bandwidth is crucial for good performance. A phenomenon called "physical layer capture" (PLC) can severely disrupt this fairness. PLC occurs when two devices transmit simultaneously, and the receiver can still successfully decode the packet with the stronger signal, even though a collision technically happened. This means the stronger sender gets its packet through, while the weaker sender might experience a collision, leading to retransmissions and a longer wait for its next turn. Over time, this disproportionately benefits stronger senders, leaving weaker ones with significantly less throughput. Existing simulation tools often don't accurately model this effect, and the paper aims to experimentally confirm its impact and explore practical solutions.

Methods

The researchers used the ORBIT indoor wireless testbed, a grid of 64 wireless nodes, to conduct their experiments. They employed Atheros-based Wi-Fi cards that allowed for software control of various MAC layer parameters.

Key methods include:

  • Experimental Setup: A controlled 8x8 grid testbed with nodes equipped with Atheros Wi-Fi cards was used. This setup allowed for precise timing measurements and software manipulation of network parameters.
  • Capture Detection: They used a technique involving multiple sniffers placed near senders. By merging the timestamps from these sniffers, they could reconstruct a global timeline of packet transmissions and receptions, allowing them to identify collisions and subsequent captures.
  • Fairness Evaluation: The paper quantifies unfairness by measuring the throughput of competing flows using the iperf tool.
  • Mitigation Techniques: The core of the paper involves evaluating several methods to restore fairness:
  • Transmission Power Control: Reducing the transmit power of the stronger sender.
  • MAC Retry Limit Adjustment: Limiting the number of retransmission attempts for weaker senders.
  • 802.11e EDCF Parameter Tuning: This is a significant focus, exploring:
  • CWmin (Contention Window Minimum): Adjusting the initial backoff window size.
  • TxOp (Transmission Opportunity): Allowing a station to send multiple packets in a single channel access opportunity.
  • AIFS (Arbitration Inter-frame Space): Modifying the mandatory spacing between frames.
  • Joint Adaptation: Finally, they proposed and evaluated a heuristic approach combining TxOp and AIFS adjustments for multiple flows.

Datasets

Not discussed in the paper. The experiments were conducted on a custom-built testbed (ORBIT) using standard 802.11b traffic generated by iperf.

Key Insights and Ideas

  • Physical Layer Capture is Real and Impacts Fairness: The paper provides experimental evidence that PLC significantly reduces throughput fairness in off-the-shelf 802.11 networks, a phenomenon often overlooked or poorly modeled.
  • EDCF Parameters Offer Fine-Grained Control: The 802.11e Enhanced Distributed Coordination Function (EDCF) parameters (TxOp and AIFS) proved to be the most effective tools for fine-tuning fairness, offering more granular control than transmission power or retry limits.
  • TxOp and AIFS are Promising for Fairness Restoration: Specifically, increasing the TxOp for weaker senders and adjusting AIFS (often by increasing it for stronger senders) showed significant improvements in fairness.
  • Limitations of Power Control: While reducing transmit power can help, its effectiveness is limited by discrete power levels available on hardware and the lack of per-packet adaptation.
  • Retry Limit Adjustment is Application-Dependent: Reducing retries for weaker senders can improve their throughput but is only suitable for applications tolerant to packet loss.

Key Results

  • Quantified Unfairness: The paper shows that PLC can lead to throughput imbalances as high as 5x between stronger and weaker senders.
  • Transmission Power Control: Reduced the gap between flow throughputs and signal strength differences, but was limited by hardware capabilities.
  • MAC Retry Limit: Decreasing the retry limit for the weaker sender improved its throughput, but this is only viable for loss-tolerant applications.
  • EDCF Parameter Tuning:
    • CWmin: Adjusting CWmin helped, but hardware limitations (rounding to powers of two) restricted fine-tuning.
  • TxOp: Increasing TxOp for weaker senders showed significant improvements, restoring fairness for 1024-byte packets when set to 2-3 packets per burst.
  • AIFS: Increasing AIFS for stronger senders also helped achieve fairness by reducing collisions.
  • Heuristic Approach: A two-step heuristic combining TxOp and AIFS adjustments for multiple flows improved the Jain's fairness index by approximately 25% compared to the default settings.

Limitations and Future Work

  • Hardware Limitations: The granularity of control for transmission power and CWmin is limited by the underlying hardware.
  • Fixed PHY Rate: Experiments were conducted at a fixed PHY rate. The impact of auto-rate selection on fairness in the presence of PLC was not studied but is identified as a future research area.
  • Future Work:
    • Develop efficient algorithms for capture detection and fairness restoration using distributed sniffers and receiver feedback.
  • Investigate the effectiveness of RTS/CTS mechanisms in mitigating PLC-induced unfairness.
  • Study the impact of variable bit-rate environments and auto-rate selection.
  • Explore more sophisticated joint adaptation strategies for all parameters.

Summary updated

BibTeX
@inproceedings{ganu2006methods,
  author = {Sachin Ganu and Kishore Ramachandran and Marco Gruteser and Ivan Seskar and Jing Deng},
  title = {{Methods for restoring MAC layer fairness in IEEE 802.11 networks with physical layer capture}},
  booktitle = {REALMAN@MobiHoc},
  year = {2006},
  doi = {10.1145/1132983.1132986},
}
Figure