Conclusions

At present, new issues of high-speed network research emerge from the convergence of wireless and wired next generation networks and the planned deployment of new integrated multi-media and Web services. In our paper we have sketched the concept of a service architecture where adaptive applications are running above a distributed service management and QoS-control layer on top of the transport and network layers with their corresponding resource allocation and control as well as mobility-management functionalities. We have described a LINUX implementation of our concept in an IEEE802.11b compatible wireless LAN with Mobile IP as network layer using a unified programming model.
Furthermore, we have investigated the efficiency of the data transport by the TCP and UDP protocols. By appropriate measurement tools we have illustrated the impacts of the changing transmission quality, the resulting error recovery of the wireless data link layer and the roaming between different basic service areas on the dynamics of TCP flow control and on the resource-reservation process taking into account realistic emission patterns of the IP frames. Moreover, we have identified the most effective and sensitive control information of the adaptation process.
Finally, we want to point out some important issues regarding the interworking between resource reservation, QoS-control, mobility- and security-management that require an improved, more efficient solution in the near future. From our experiments, one can conclude that an interworking between the TCP flow-control and the behavior of the wireless DLC layer should be organized to improve the throughput and delay characteristics of the data flows while moving and roaming, e.g. by exchanging management and control information that optimizes an ECN-based TCP flow-control in a way similar to the ABR-scheme in ATM (cf. [26]). The use of improved TCP variants, e.g. those that freeze and recover the congestion-window state after a handoff, that use TCP control block statistics, that mimic a loss-less PDU transfer at the data link layer like snoop or that distinguish different sources of packet loss may be another alternative (cf. [6], [29]). Furthermore, we have demonstrated, despite all technical shortcomings of our implementation, that micro-mobility is not supported in an adequate manner by the current functionality of Mobile IP. Cellular IP, route and address caching techniques, improved movement detection by eager or hinted cell switching and fast or two-phase handover schemes may partially resolve the observed difficulties (cf. [8], [14], [16]).
In conclusion, we think that the realization of our concept and the measurements clearly reveal the potential and drawbacks of the mobile-aware approach and provide new insight on the relationship of the signal-to-interference ratio, the delay-loss characteristic of flows and the TCP behavior as well as the adaptation processes. They may also be used to develop improved handoff and TCP flow-control mechanisms as well as improved mobile-aware applications for an efficient transport of real-time multi-media and interactive Web services in next generation wireless networks.

2002-02-21