SLA driven performance optimization in cloud computing

Abstract

KEYWORDS: Service Level Agreement; Cloud Economics; Pricing Model; Agent-Based Cloud Computing; Dynamic Assignment Algorithm. The main idea behind cloud computing technology is to implement “Everything-as-a-service”. In realizing such a service there are mainly two issues: technology and business. The focus in technology is towards adaptation and implementation, IT strategy/policy including security, and other related aspects. On the other hand, in the business category, cloud computing economics, regulatory issues are among the main concerns. Substantial research has been undertaken to address the technology issues; however, if cloud computing is to achieve its true potential, there needs to be a clear understanding and subsequent development in the various aspects of cloud computing economics both from the perspective of providers and the consumers of the technology. There is a strong correlation between economics and the delivered Quality of Service (QoS). For good economics, therefore, it is imperative that a Service Level Agreement (SLA) is real ized between cloud consumers and cloud providers. Service Level Agreements (SLAs) play a key role in translating IT effectiveness into measurable business value. Contemporary cloud SLA mechanisms are typically limited to cost-performance trade-offs, where both service providers’ revenue and consumers’ QoS are intended to be optimized. In fact, these mecha nisms are trivial and do not take several significant factors such as trust, risk, power into ac count. As an alternative therefore, more involved economic models incorporating these factors are required. These models, among other things, should include parameters to compare per formance repercussions between alternative configurations. Furthermore, such models should employ business-related terms that can be translated to service and infrastructure parameters during development and deployment of services. Effectively doing this requires optimization of several factors through a run-time monitoring mechanism that includes analysis of histori cal usage patterns and prediction of future events. Such management in clouds is non-trivial owing to the ever growing complexity and inherent variability in services.