Applying active network adaptability to wireless networks

The IP-based Internet, although wildly successful, is limited in its ability to evolve and adapt, in particular at the network layer. Mobile/wireless networking is an important emerging area in which adaptivity and evolvability is likely to be especially important due in part to the widely varying nature of the underlying communication channels themselves. We believe that active networking (AN) enables valuable adaptivity that existing technologies currently lack. This is because AN enables highly flexible packet functionality, on-the-fly protocol deployment, even on a packet-by-packet granularity, and cost-effective network expansion. Because adaptivity and evolvability is at a premium, we have chosen to test our belief in the mobile/wireless networking domain using three case studies: Mobile IP, ad hoc routing, and TCP over wireless. In our work, we demonstrate AN’s adaptivity by developing a series of designs, simulation studies, and working prototypes. Mobile IP is a protocol that supports mobility within the existing IP architecture by separating naming and addressing. While its design fits the conventional architecture and is feasible in current networks, Mobile IP exemplifies the inability of current networks to evolve effectively. Using Active Packet evolution and Update evolution techniques, we show how to deploy the new protocol and to evolve networks to support Mobile IP. Ad hoc networks are infrastructureless networks in which hosts are typically mobile and must act as routers. Mobility makes routing hard because the state of links changes frequently and routing heterogeneity is likely. We show how AN can provide useful routing adaptation to host mobility, in addition to routing evolution. In the last case study, we address the performance degradation of TCP over lossy links. TCP’s congestion control may cause under-utilization of bandwidth in wireless networks. We demonstrate AN’s adaptation to changing link conditions. Furthermore, taking advantage of flexible cross-layer interactions, we show AN’s ability to adapt to changes in TCP flow information. We show that active packets are especially useful in this context because they are extremely agile and allow adaptation on a packet-by-packet basis.