Adaptive inverse multiplexing for wide-area wireless networks

The limited bandwidth of current wide-area wireless access networks (WWANs) is often insufficient for demanding applications, such as streaming audio or video, data mining applications, or high-resolution imaging. Inverse multiplexing is a standard application-transparent method used to provide higher end-to-end bandwidth by splitting traffic across multiple physical links, creating a single logical channel. While commonly used in ISDN and analog dialup installations, current implementations are designed for private links with stable channel characteristics. Unfortunately, most WWAN technologies use shared channels with highly variable link characteristics, including bandwidth, latency, and loss rates. This paper presents an adaptive inverse multiplexing scheme for WWAN environments, termed link quality balancing, which uses relative performance metrics to adjust traffic scheduling across bundled links. By exchanging loss rate information, we compute relative short-term available bandwidths for each link. We discuss the challenges of adaptation in a WWAN network, CDPD in particular, and present performance measurements of our current implementation of wide-area multi-link PPP (WAMP) for CDPD modems under both constant bit rate (CBR) and TCP loads.