When considering communities mediated by computer network, the efficient usage of limited resource in network shared by the people is a big issue. For instance, in the Internet and intranets, which are shared by users who use various types of application programs, conflicts among their demands are considered a serious problem.
Consider the case of desktop conferencing system. A single user of such a multimedia communication tool can make use of a large amount of bandwidth. In a shared network environment, these demands conflict each other, and the arbitration among them is generally hard because of the vast variety of user preference and application types. To achieve efficient usage of limited network resource, the method to synthesize the preference of the users is required.
Our approach is to introduce price adjustment mechanisms of market which have been studied in microeconomics, where the preference of humans are modeled as utility to achieve efficient usage of limited resources. We apply these mechanisms to network systems, and validate this approach by the implementation to the actual application softwares.
To achieve the above objective, the preference of various users to various multimedia communication links has to be included in a model. Our approach assumes that users evaluate the set of services provided by application programs instead of raw network performance. Economic agents, consumer that represent the users' preference and producer that represent the performance of application programs, trade resource and achieve efficient allocation.
So far, we have applied a market-based allocation algorithm used in WALRAS, a market-oriented programming environment, to FreeWalk, a desktop conferencing system which use a virtual three-dimensional space. We created a market model for the environment (Figure 11), and performed analyses both by simulation and implementation.
The model reflects the characteristics of FreeWalk that the preference of users for individual communication links varies based on their relative locations in shared virtual space and changes dynamically along with their movement, by incorporating two temporal categories, ``current'' and ``future,'' and allowing users to trade the right to use network resource through time .
To allocate network resources, the demand for which changes dynamically, the speed of calculation is critical as well as the quality. We implemented the above mechanism to FreeWalk and analyzed several elements that affect the calculation time. Through the analysis, we found out two types of tradeoff, spatial tradeoff between computation and communication, which resides in the distribution of calculation elements, and temporal tradeoff between accuracy and reactiveness . Figure 12 shows the temporal tradeoff in an implementation of QoS control mechanism for FreeWalk. The horizontal axis is the time of market calculation, and the vertical axis represents deviation between the result of calculation and the environment that dynamically changes. This figure proves that (1) temporal trade off exists in the actual environment, and that (2) market-based algorithm can achieve efficient usage of network in a highly dynamic environment.
As the next step, we apply our approach to the more general desktop conferencing environments than FreeWalk. Concretely, we implement a system that controls the resource allocation for desktop conferencing in an intranet of an office which includes multiple network segments. We aim to implement and evaluate a general platform, where users input their preference and the system allocate resource to maximize the utility of the community.
There are an enormous number of networks shared by human community, such as the Internet and various intranets, and the performance and the administrative policy varies from network to network. We aim to provide formal methods to create market models that fit to each community, by implementing those mechanisms to actual applications and evaluating them.
Servers TD> A>