On Channel and Transport Layer aware Scheduling and Congestion Control in Wireless Networks

In this thesis, we consider resource allocation schemes for infrastructure-based multipoint-topoint wireless networks like IEEE 802.16 networks and infrastructure-less ad-hoc networks. In the multipoint-to-point networks, we propose channel and Transport layer aware uplink scheduling schemes for both real-time and best effort services, whereas in ad-hoc networks, we propose channel aware congestion control schemes for best effort services. We begin with the performance evaluation of IEEE 802.16 networks by conducting experiments in the current deployed IEEE 802.16 networks of a leading telecom operator in India. We observe that (i) the throughput of Transmission Control Protocol (TCP) based application is poor as compared to that of User Datagram Protocol (UDP) based application and (ii) TCPbased application suffers in the presence of simultaneous UDP-based application. The Weightbased (WB) scheduler employed in this service provider network does not provide any delay and throughput guarantee to these applications. This motivates us to further investigate scheduling schemes specific to real-time and best effort services (TCP-based applications). For real-time services, we propose a variant of deficit round robin scheduler, which attempts to schedule uplink flows based on deadlines of their packets and at the same time attempts to exploit the channel condition opportunistically. We call this as “Opportunistic Deficit Round Robin (O-DRR)” scheduler. It is a polling based and low complexity scheduling scheme which operates only on flow level information received from the users. Our key contribution is to design scheduling mechanism that decides how many slots should be assigned to each user based on its requirement and channel condition. Since O-DRR employs no admission control and does not maintain packet level information, strict delay guarantees cannot be provided to each user. Rather, O-DRR reduces deadline violations compared to the Round Robin (RR) scheduler by taking the deadlines of Head of the Line (HoL) packets into account. For fairness, the O-DRR scheduler employs the concept of deficit counter similar to that in Deficit Round Robin (DRR) scheme. We demonstrate that O-DRR achieves lesser packet drops and higher v