Quality-of-Service Measurements with Model-Based Management for Networked Applications

Distributed applications are evolving towards compositions of modular software components with user interfaces based on web browsers. Each of these components provides welldefined services that interact with other components via network. The increase in the complexity of distribution makes it more difficult to manage the end-to-end Quality-of-Service (QoS). The challenge derives in part since different management scopes of network and computing domains need to interact. We address two needs of a management system deployed to diagnose QoS degradation. First, to measure the performance of applications, it needs a low-overhead, scalable system for measuring software components. Second, the performance management system must monitor selected measurements, diagnose QoS degradation, adapt to the environment and integrate with network management systems. We extend our Distributed Measurement System (DMS) into browser-based Java applets to deliver low-overhead and pertinent performance information to a management system. A model-based reasoning engine equipped with “generic” application models uses these measurements to diagnose QoS trends. These models incorporate the notion of composite transactions and the organization of distributed components. We demonstrate QoS monitoring using this architecture on a typical, component-based distributed application deployed in a wide area network.