Economic paradigms for information systems and networks
Jakka Sairamesh · 1997
In this thesis, we propose novel methodologies based on economic models and market mechanisms to provide Quality of Service (QoS) to competing user groups or user classes in information networks and systems. We first consider an economy of a packet network where resources are priced by network service providers. User classes, via economic agents, compete for network resources (bandwidth and buffers), and purchase them to satisfy their diverse QoS needs (such as packet loss, delay and thruput). Network providers set prices based on demand and available supply. Using such economic models, we provide a new definition for QoS provisioning, where we obtain the set of optimal resource allocations (Pareto optimal) to provide QoS guarantees to competing user classes. We decentralize the control of resources and services such as scheduling, admission and routing. In addition, we propose novel session routing mechanisms, which are performed by the agents. In our network economy, prices are set based on demands from agents, who represent user classes. Users within a class share resources, and in order to account for them, we propose novel charging mechanisms. Our economic models can be applied to QoS provisioning and QoS based charging/accounting in networks such as ATM and the upcoming Integrated Services Internet. In addition, our economic models can be applied to environments such as private or campus networks and systems, where fictitious money can be used for allocating quotas among users or user groups. We then consider QoS provisioning and resource allocation issues in distributed information systems, where the resources are server processing rate and memory. Users of applications (such as multimedia, transaction processing and database), in order to satisfy their QoS needs, compete for resources from servers of the information system. We consider an economy of such a system, and we show Pareto optimal allocation of resources among competing economic agents, who represent users classes (such as transaction classes). The QoS requirements are average response time and thruput. We show novel static and dynamic session routing algorithms by agents, where control is decentralized. In a large economy, consumers have to find suppliers, and likewise, suppliers have to advertise in order to attract consumers. We present three architectures using Internet Domain name Services to promote interaction between consumers and suppliers in an economy. This interaction is done through bulletin boards where suppliers advertise their services, and consumers search for the right set of suppliers. We design and implement the three architectures, and we compare the architectures based on performance (average consumer search response time), scalability, effective caching/replication and security.