CSMA with implicit scheduling through state-keeping: A distributed MAC framework for QoS in broadcast LANs.

摘要

Channel access fairness and efficiency in capacity utilization are the two main objectives for Quality of Service (QoS) specific to Medium Access Control (MAC) protocols in computer networks. For bursty and unpredictable traffic in networks, fairness and efficiency involve a mutual tradeoff with the currently popular QoS mechanisms. We propose a QoS MAC framework for carrier sensing multiple access (CSMA) networks, that achieves fairness with improved efficiency through extensive state-keeping based on the MAC evolution. This CSMA with Implicit Scheduling through State-keeping (CSMA/ISS) framework involves the tracking of traffic arrival at active nodes, the nodes that need channel access frequently. It also involves implicit channel access grants to different active nodes according to their estimated queue backlogs and the fair scheduling requirements. These methods save channel capacity that may otherwise be required for disseminating the access requirements of various nodes, and their access rights according to fairness rules. A static, hierarchical, and weighted fair access scheme is designed in CSMA/ISS by allowing repeated rounds of access that are weighted fairly according to requirements. Weighted fairness across classes is achieved by invoking channel access for each traffic class in a round as many times as its weight. Within each class, all active nodes are allowed equal access through in-order channel access based on a looped list of active nodes. Although CSMA/ISS is proposed as a distributed control framework for efficiency, it may also be employed in central control protocols. It may also be adapted to different types of CSMA networks, both wireless and wired, by an appropriate choice of the underlying classical access mechanism. The CSMA/ISS framework was modeled and simulated as a QoS capable MAC protocol for a wired fully connected local network environment. We present the CSMA/ISS framework, the example implementation, and the results of performance evaluation of the example implementation. Significant performance improvements were observed, and the memory and processing trade-off was found to be low to moderate.