3rd International Conference on Information and Computing Science (ICIC 2010), Wuxi, China Towards Measuring the Degree of Fulfillment of Service Level Agreements Frank Schulz SAP Research Karlsruhe SAP AG Karlsruhe, Germany frank.schulz@sap.com Abstract— In service oriented architectures (SOA), the non- functional properties of services have been recognized to be highly important in addition to the functionality of services as a means to differentiate services according to quality considerations. Service level agreements (SLAs) are formalized contracts between service providers and service consumers that are used to define quality of service (QoS) properties. The violation of an SLA by the service provider typically results in a penalty to compensate the service consumer. In order to avoid such situations, the service provider needs to recognize critical service instances and to take appropriate countermeasures before a violation happens. Therefore a measure for quantifying the danger of SLA violation is needed as part of a service level management system. This paper proposes a concept for the definition and evaluation of such a metric that takes into account the underlying structure of the SLA as well as the available options for monitoring service quality parameters. Hence it becomes possible to obtain detailed information of the status of service fulfillment at runtime and to identify critical service instances. The methodology is exemplified with the availability property. Keywords- service oriented architecture, service level agreement; quality of service; key performance indicator; monitoring I. INTRODUCTION Service oriented architectures have emerged as a highly successful approach to build modular software systems [1]. In their realization with Web Services, they form the conceptual basis for many e-commerce solutions and for innovative approaches to deliver software applications to customers, like Software-as-a-Service (SaaS) and On- Demand computing. Similarly, the combination of existing services to new value-added services leads to the formation of agile service networks that represent new forms of collaboration between companies or institutions. In business contexts, where consumers of such services rely on them to operate their own business, the quality of the service offers becomes vital [2]. Service Level Agreements are the common approach to define levels of service quality. An SLA constitutes a formal agreement negotiated between a service provider and a service consumer about the consumption of a service. It contains a definition of the service (e.g. by linking to a WSDL description of the service), the duties of the involved parties, the agreed guarantees, and optionally additional information like penalties in case of failure to comply with the defined service level objectives. Before accepting a Service Level Agreement, the service provider has to ensure that sufficient capabilities for fulfilling the agreement will be available. As soon as a Service Level Agreement has been negotiated and accepted and the corresponding service is operational, the service provider needs to guarantee that service delivery complies with the agreement. Therefore a continuous monitoring of all services at runtime is needed and critical services have to be discovered quickly. The management of a service provider’s infrastructure with the help of SLAs is an active research area [3,4]. Currently there are two main approaches for the management of service operations at runtime. One direction is based on the monitoring of SLA violations [5]. In case of a SLA violation, strategies like the provisioning of additional resources can be applied in order to remedy the situation. This approach is based on the binary information whether a given SLA is fulfilled or violated. The second direction uses low-level monitoring for assuring proper operation of system resources [6]. It results in detailed key performance indicators (KPIs); however the connection to consumer requirements via SLAs is missing. This paper aims at filling this gap. By quantifying the degree of SLA fulfillment, it becomes possible to obtain a fine-granular status of all service instances and their corresponding consumer relations. Thus, the service landscape can be closely monitored on the level of SLAs, and critical service instances with imminent danger of SLA violation can be identified. This information is needed as the basis for further decisions within the service level management. Possible actions include the provisioning of additional resources or the cancellation of some service instances with low penalty. The main contributions of this paper are (1) a classification of key performance indicators with respect to aggregation properties, (2) a definition of the degree of fulfillment based on aggregated KPIs and (3) a method for combining these atomic data according to the logical structure of the SLA. Section 2 discusses related work. In section 3, technical foundations for our approach are presented, and section 4 contains the details of the new concept. Section 5 concludes with a summary and outlook.