Efficient Data Transport in Wireless Overlay Networks

Wireless data access for nomadic users is predicted to be a major growth direction for the future Internet. However, no single existing wireless technology fits all user requirements at all times. Instead, several overlaying wireless networks can provide best possible data delivery service. Nomadic users can run interactive, conversational, streaming, and background applications that rely on end-to-end transport protocols to communicate over unreliable wireless links. Achieving efficient data transport in wireless overlay networks implies meeting Quality of Service requirements of applications while preserving radio resources, battery power, and friendliness to other flows on the Internet. Events such as delay spikes, bandwidth oscillation, and connectivity outages are difficult to prevent in the heterogeneous and dynamic wireless environment. For instance, delay spikes can be caused by handovers, higher priority voice calls, and link layer retransmissions. Furthermore, link characteristics can change by an order of magnitude when a handover is executed between overlay networks. Such disruptive events can cause delivery of duplicate, stale, aborted data, and low utilization of the wireless link. Achieving efficient data transport in this environment demands coordinated efforts from the link layer and from end-to-end transport protocols. In this dissertation, existing and emerging wireless networks are examined through measurements and simulations. We paid special attention to the models used in simulations. We studied end-to-end transport of real-time and non-real-time data. For non-real-time data, TCP is a highly suitable transport protocol when profiled with state-of-the-art features and when its robustness to delay spikes is improved. We measured the response of different TCP variants to delay spikes and developed mechanisms to alleviate negative effects of spurious timeouts in TCP. Delay spikes in the network can often make real-time data useless to the receiver. For streaming and conversational traffic we suggested using a transport protocol that incorporates an explicit lifetime into packet headers. The Lifetime Packet Discard eliminates stale and duplicate data delivery over the wireless link that preserves radio resources and battery power of wireless users. An inter-system handover can cause an abrupt change in the link bandwidth and latency. It is hard for end-to-end congestion control to adapt promptly to such changes. This is especially a concern for slowly responsive congestion control algorithms, such as TCP-Friendly Rate Control (TFRC), that are designed to provide a smooth transmission rate for real-time applications and therefore are less responsive to changes in network conditions than TCP. We measured the performance of TFRC and TCP flows during vertical handovers in overlay networks in a testbed and using a simulator. Overbuffering and an explicit handover notification are shown to improve transport performance during vertical handovers. Computing Reviews (1998)