Achieving high throughput and low delay in wireless networks
For wireless networks, although throughput optimization is heavily studied, it has been shown that known throughput-optimal algorithms either incur high-complexity (e.g. max-weight algorithms) or exhibit poor delay-performance (e.g. CSMA algorithms). Further, their delay performance is often very difficult to characterize. Hence, designing low-complexity algorithms that provide both good throughput- and provable delay-performance in wireless networks is an open and challenging research problem.^ In this dissertation, we tackle this challenging research problem. The main idea behind the algorithm design is to schedule the link transmissions based on the desired rate allocation. We refer to this class of scheduling algorithms as the "rate-based scheduling algorithm." The main advantages of such algorithms are that it can be combined with the window-based flow control to directly control the packets in the network, and the per-flow delay performance can be more precisely characterized. Specifically, for fixed-route flows, we develop a novel stochastic dominance approach for rate-based scheduling algorithms, and the derived per-flow delay bound is order-optimal with respect to the number of hops. For multi-path flows, we utilize the virtual-circuit approach to extend the analysis and propose two queue merging rules, which lead to improved delay bounds that account for the statistical multiplexing gains.^ We then develop distributed rate-based scheduling algorithms to achieve high throughput. Specifically, for multi-path multi-hop traffic, our proposed algorithms can achieve a provable portion of the system capacity with low-complexity. To achieve optimal system capacity, in the latter part of the thesis, we propose a new Virtual-Multi-Channel (VMC-) CSMA algorithm, which combines the idea of rate-based scheduling algorithm with CSMA-like algorithms to achieve optimal system capacity and low delay. The key idea is to utilize a virtual multi-channel system such that the optimal rate allocation can be computed implicitly by CSMA-like algorithms. Hence, by serving the schedules in different virtual channel randomly, the VMC-CSMA algorithm behaves like a rate-based scheduling algorithm. Under a single-hop utility-maximization setting, we show that VMC-CSMA can approach arbitrarily close-to-optimal system utility with the computation complexity increasing logarithmically with the network size.^
Xiaojun Lin, Purdue University.
Engineering, Electronics and Electrical