Scalable network architectures, protocols and measurements for adaptive quality of service.

摘要

New IP services and applications have diverse and stringent bandwidth and quality of service requirements. It is difficult and costly to predict these requirements and add sufficient capacity to provide reliable and high quality service. This thesis is broadly concerned with scalable and efficient architectures for delivering Internet applications reliably and with high quality. The work consists of two parts.; We have developed a flexible, scalable service framework to utilize existing network capacity efficiently. The framework is based on conditioning the network to provide multiple services with certain quality of service expectations even under high demand, using short-term, dynamic configuration of network resources for better response to user demand and more efficient resource usage. To support this framework, we develop a Resource Negotiation and Pricing Protocol (RNAP) to enable the user and the network (or two peer networks) to dynamically negotiate services, and to formulate and communicate prices and charges.; As part of the service framework, we also propose a pricing model in which services are priced based on QoS (resources consumed) and user willingness-to-pay. The model also motivates rate and service adaptation by applications with elastic demand through congestion-sensitive pricing of certain services, while providing more expensive services with static pricing for non-elastic users. We describe two mechanisms to develop a congestion-sensitive price component, one based on a tâtonnement process, and the other on a second-price, multiple-bid auction. We also describe a user rate-adaptation model in response to price changes, based on maximization of the user-perceived benefit.; We have demonstrated the functionality of RNAP on a testbed network, and have shown that the framework can achieve high utilization and control congestion, while maintaining a stable price. We also present simulation results to show that our proposed framework achieves a lower connection blocking rate, higher user satisfaction based on user utility functions, and higher network revenue. The performance improvement increases with the number of connections. The auction-based congestion pricing mechanism is seen to provide higher network utilization than the tâtonnement-based mechanism, at the cost of higher complexity.; The second part of the thesis is a study of the performance of the Lightweight Directory Access Protocol (LDAP). LDAP is potentially useful for the management of network resources, and the administration of traffic-handling and pricing policies. We describe a benchmark tool to study LDAP performance in a dynamic environment. The tool provides a detailed profile of the latency and throughput contributions of system components. We report measured performance using a LDAP schema proposed for administration of Service Level Specifications in a differentiated network. In most cases, the connection management latency increases sharply at high loads and thus dominate the response time. The TCP Nagle algorithm is found to introduce a very large additional latency, and it appears beneficial to disable it in the LDAP server. The CPU capability is found to be significant in limiting the performance of the LDAP server, and for larger directories, which cannot be kept in memory, data transfer from the disk also plays a major role. The scaling of server performance with the number of directory entries is determined by the increase in back-end search latency, and scaling with directory entry size is limited by the front-end encoding of search results, and, for out-of-memory directories, by the disk access latency.