Over the years, the layering principle, which segments network functions and specifies standard interfaces between layers, has served the networking community quite well. The standardization of functional interactions between the layers has, to some extent, allowed developers to work independently of each other. This has led to the development of multiple sources for protocols and new market-driven approaches to their use and integration.
Three relatively recent phenomena have affected the way protocol stacks will have to be designed. First, protocol stacks are being used in new ways. Two examples are "best effort" IP over ATM stacks that can guarantee service quality and voice, which demands some quality guarantees, over best effort IP. Second, concepts such as quality of service guarantees and energy conservation, which can be engineered in multiple ways at various layers, are emerging. Finally, concepts like software controlled radios and active networks that can be used to redefine networking functions on the fly are being developed. These three sets of developments taken together stimulate the need for better understanding of the interaction between the layers. IPSI, ATR, Mitsubishi, and NEC are working on these developments. Representatives of Hughes, a U.S. company, described its interests in this topic as well.
Mitsubishi researchers reported that their concern for reducing the size and weight of hand sets compels the exploration of cross layer opportunities for effecting efficient communications. At NEC, interest in provisioning QoS, has led to a call for "definitions that work" and enhanced understanding of the functional relationships between levels in order to effect system optimization. At NEC, Dr. Hori articulated the need "to mix the software/system issues with hardware." For instance, NEC has an interest in a system on a chip with integrated antennas.
The objectives of ATR's Adaptive QoS Management project are to develop adaptive ways in which network parameters and resources are controlled in response to changing QoS demands in multimedia applications. The focus at ATR was on controlling the compression and transmission parameters of video streams (bit-rate, frame-rate, etc.) through negotiation among network agents that correspond to different user streams. Eventually these negotiations will include allocation of additional network resources (such as buffer memory, bandwidth, etc.). The investigators do realize that end-to-end QoS performance depends on the lower-layer (link) QoS parameters and intend to include this coupling in their methods.
The visit at IPSI highlighted the issues in wireless networking that are often ignored by the wireless community, namely the need to address the design of upper-layer protocols and their effects on link operation. However, the visit verified that an integrated approach to networking that bridges the gaps among layers is still needed, but not yet in place.
A number of companies, especially those that were building prototypes, were keenly aware of the need to develop a comprehensive understanding of design options regardless of the layer (of the system protocol stack) that the function traditionally resides in. The most pressing problems identified included (1) "QoS definitions that work," (2) better understanding of the functional relationships between levels, (3) interlayer interactions, and (4) ways to mix software, systems, and hardware issues. More focused research is required in order to understand the (1) effect of lower-layer (link) QoS characteristics on end-to-end QoS performance and (2) the effect of upper-layer protocols on link operation. This research is necessary in order to support more sophisticated negotiation between applications and networks.
From a larger perspective, the payoff from adopting integrated approaches may not be worth the cost of deconstructing the layered approach to protocol design. Indeed, layering protocol stacks has served well in the past. Yet, developments that motivate new research include the following: (1) the notion of a link layer is much softer in the wireless environment than in a wire-line environment, (2) issues such as quality of service guarantees and energy conservation can be engineered in multiple ways at various layers and, finally, (3) new uses for protocol stacks. These developments require better understood interactions between the layers of a protocol stack as well as between different protocol stacks. Research on this topic is likely to further benefit from the availability of software controlled radios and the deployment of active networks that can be used to redefine networking functions on the fly.
Japanese companies seemed to display the highest sensitivity to this issue. Perhaps this is consistent with their early prototyping efforts. U.S. and European companies were on par with each other, but did not seem as advanced as the Japanese.