In today’s IT service market customers urge providers to grant guarantees for quality of service (QoS) which are laid down in Service Level Agreements (SLAs). To satisfy customers and to avoid penalties, service providers have to ensure that the agreed SLAs are met. Therefore, it is necessary to be able to effectively deal with resource failures which could endanger the SLAs by affecting the provided services. The effort for recovering from failures should be selected corresponding to the expected SLA violation costs. In this paper we present a framework to automatically determine the impact of resource failures with respect to services and service level agreements. We achieve this by monitoring the service quality from inside and outside the service provider and also by incorporating information about the current and expected future service usage. The expected costs of the resource failures are assessed to select an appropriate recovery alternative. Besides this short term perspective the impact analysis can also be employed to identify critical resources and to improve the service provisioning. 1

service management, auto-discovery, service models, Internet services © Copyright Hewlett-Packard Company 1999 Auto-discovery is one of the key technologies that enables management systems to be quickly customized to the environments that they are intended to manage. As Internet services have grown in complexity in recent years, it is no longer sufficient to monitor and manage these services in isolation. Instead, it is critical that management systems discover dependencies that exist among Internet services, and use this knowledge for correlation of measurement results, so as to determine the root-causes of problems. While most existing management systems have focussed on discovery of host, servers, and network elements in isolation, in this paper, we describe auto-discovery techniques that discover relationships among services. Since new Internet services and service elements are being deployed at a rapid pace, it is essential that the discovery methodologies be implemented in an extensible manner, so that new discovery capabilities can be incrementally added to the management system. In this paper, we present an extensible architecture for auto-discovery and describe a prototype implementation of this architecture and associated auto-discovery techniques. We also highlight experiences from applying these techniques to discover real-world ISP systems. Although described in the context of ISP systems, the concepts described in this paper are applicable for the discovery of services and interservice relationships in enterprise systems as well. 1