Date of Award
Doctor of Philosophy in Electrical Engineering - (Ph.D.)
Electrical and Computer Engineering
Modern communication networks are often designed for diverse applications, such as voice, data and video. Packet-switching is often adapted in today’s networks to transmit multiple types of traffic. In packet-switching networks, network performance is directly affected by how the networks handle their packets. This work addresses the packet-handling issues from the following two aspects: Quality of Service (QoS) and network coding.
QoS has been a well-addressed issue in the study of IP-based networks. Generally, nodes in a network need to be informed of the state of each communication link in order to make intelligent decisions to route packets according to their QoS demands. The link state can, however, change rapidly in a network; therefore, nodes would have to receive frequent link state updates in order to maintain the latest link state information at all times. Frequent link state updating is resource-consuming and hence impractical in network design. Therefore, there is a trade-off between the link state updating frequency and the QoS routing performance. It is necessary to design a link state update algorithm that utilizes less frequent link state updates to achieve a high degree of satisfaction in QoS performance. The first part of this work addresses this link state update problem and provides two solutions: ROSE and Smart Packet Marking. ROSE is a class-based link state update algorithm, in which the class boundaries are designed based on the statistical data of users’ QoS requests. By doing so, link state update is triggered only when certain necessary conditions are met. For example, if the available bandwidth of a link is fluctuating within a range that is higher than the highest possible bandwidth request, there is no need to update the state of this link. Smart Packet Marking utilizes a similar concept like ROSE, except that the link state information is carried in the probing packet sent in conjunction with each connection request instead of through link state updates.
The second part of this work addresses the packet-handling issue by means of network coding. Instead of the traditional store-and-forward approach, network coding allows intermediate nodes in a multi-hop path to code multiple packets into one in order to reduce bandwidth consumption. The coded packet can later be decoded by its recipients to retrieve the original plain packet. Network coding is found to be beneficial in many network applications. This dissertation makes contributions in network coding in two areas: peer-to-peer file sharing and wireless ad-hoc networks. The benefit of network coding in peer-to-peer file sharing networks is analyzed, and a network coding algorithm – Downloader-Initiated Random Linear Network Coding (DRLNC) – is proposed. DLRNC shifts the coding decision from the seeders to the leechers; by doing so it solves the “collision” problem without increasing the field size. In wireless network coding, this work addresses the implementation difficulty pertaining to MAC layer scheduling. To achieve the ideal network coding gain in wireless networks, it requires perfect MAC layer scheduling. This dissertation first provides an algorithm to solve the ideal-case MAC layer scheduling problem. Since the ideal MAC layer schedule is often difficult to realize, a practical approach is then proposed to increase the network coding performance by modifying the ACK packets in the 802.11 MAC.
Wang, Nan, "Efficient packet delivery in modern communication networks" (2011). Dissertations. 282.