Modeling quantitative requirements in SLAs with network calculus

Abstract

When planning Service-Oriented Architectures requirements declared in Service Level Agreements (SLAs) have to be considered. SLAs cover functional as well as quantitative requirements like load levels, services rates and delay times. As external factors can influence distributed systems, SLAs have to include tolerances for quantitative requirements.Early design phases of SOA use analytic models to check functional properties. However, formalization of quantitative requirements in SLAs and their validation in analytic models is still a field of research. A challenge is the description of soft deadlines and the way delay times grow under different load levels.Network Calculus system theory can give bounds on delay times in systems. It has already been used to validate hard deadlines in networks and embedded systems. For its use in SOA models, soft deadlines and other aspects derived from SLAs have to be included. This paper introduces a new method to control delay times in Network Calculus models in order to specify quantitative requirements. The basic Network Calculus concept of arrival and service curves is extended with delay curves and their relationship is discussed.