SLAD@CLOUD: um sistema de acordo de nível de serviço para computação em nuvem

Abstract

Cloud computing has become more popular. However, though the facilities and advantages offered by cloud computing, there are still obstacles to its adoption by users and limitations in the provision of services by providers. For the cloud computing can be used and be crucial to business operations of the users it is essential that they receive guarantees from suppliers in the delivery of services. Usually, these guarantees are provided through SLAs (Service Level Agreements) between providers and consumers. The current service level agreements (SLAs) offered by cloud computing providers are simple, static and predefined by providers. These SLAs have no dynamic negotiation and does not follow the elastic behavior of the cloud. Cloud services are subject to fluctuations in load and SLA violations are more likely to happen during these fluctuations. The nature of these variations are unpredictable and therefore a static SLA to withstand these conditions will not be efficient. In this sense, this work aims to present a model SLA for cloud computing called SLAd@Cloud. This model treats the complexities inherent characteristics of the cloud allowing the dynamic negotiation and renegotiation of the SLA based on the quality requirements of the application and monitoring of specific metrics and an active and integrated into the SLA to ensure the quality of services provided in the cloud computing. The model works with multiple metrics and using a system of qualification based on the methodology of pairwise comparison of Saaty, it can perform the prioritization of actions to prevent violations of specific metrics depending on the weights of these metrics. The evaluation was made through implementation of a Java prototype that interacted with the framework Cloudsim, providing interface definition and SLA hiring and managing its life cycle, even after their employment in different cloud models. The results showed gains in runtime, the financial cost and success rate of completion of the tasks, overcoming the traditional model or the model does not work with multiple metrics.