NeTS: Small: The Impact of Message Passing Complexity on QoS Provisioning in Stochastic Wireless Networks NeTS: Small: The Impact of Message Passing Complexity on QoS Provisioning in Stochastic Wireless Networks Wireless networks operate under hostile conditions, including adverse RF environments, interference, changing network topology and bursty traffic, and often exhibit multi-scale stochastic dynamics, particularly session-level dynamics (sessions come and go with finite workload) and packet-level dynamics (packets arrive in bursts and go through possibly unreliable transmissions). Quality of services (QoS) provisioning in such dynamic environments is clearly more challenging, and often requires a significant amount of message passing. Specifically, the implementation of network algorithms hinges heavily on state information exchange, and network functions are intimately tied with the complexity of message passing. Intellectual merit. The impact of message passing complexity is a fundamental yet under-explored area. Taking a transformative perspective, this project aims to pursue a systematic characterization of the impact of message passing complexity on QoS metrics in wireless networks, including effective throughput, delay and stability. Under such a common theme, the proposed research is organized into two coordinated thrusts that address session-level dynamics and packet-level dynamics, respectively. I) Vacation model for complexity in wireless scheduling. The past decade has witnessed a surge of interest in devising distributed scheduling for multi-hop wireless networks, and the complexity levels range from no message passing to constant-time complexity to exponential complexity. The corresponding effective throughput and delay, however, is not well understood when the message passing overhead is taken into account. In particular, delay characterization of wireless scheduling presents a demanding challenge and is largely open. Thus motivated, the PI will develop novel vacation models to account for the signaling complexity where vacations correspond to the durations of signaling overhead. Effective throughput (session-level stability) will be studied using the fluid approach and delay analysis will be carried out by diffusion approximation. II) Noisy feedback model for complexity in distributed rate control. The implementation of distributed rate control hinges on state information exchange, which unfortunately is often obtained using error-prone measurement mechanisms. That is to say, the feedback signals are noisy in nature in packet-level dynamics. The PI will devise noisy feedback models to account for message passing complexity in rate control algorithms using various optimization methods, and investigate stochastic stability of both single timescale algorithms and multiple timescale algorithms. The study here will provide a platform to compare different rate control algorithms in terms of complexity and robustness in packet-level dynamics. Broader impacts: A significant part of this project is to integrate research into educational activities. The PI has a long history of involving under-represented and minority students in his research. For curriculum development of advanced courses, the PI will make conscientious efforts to introduce new pedagogical methods, including guest lectures, replaying plenary talks, reading assignments of short-course lectures. Besides educational elements, the proposed research will significantly advance the understanding of the impact of message passing complexity on QoS provisioning in wireless networks, and the findings on throughput and delay will provide insight for the design of wireless applications (e.g., wireless mesh networks). In particular, the study on open problems, such as delay performance of wireless scheduling, will open up new research directions in this area.
|Effective start/end date||10/1/09 → 9/30/13|
- NSF-EHR-DUE: Division of Undergraduate Science, Engineering, & Mathem: $339,574.00
Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.