MACHINE LEARNING APPROACHES IN IMPROVING SERVICE LEVEL AGREEMENT-BASED ADMISSION CONTROL FOR A SOFTWARE-AS-A-SERVICE PROVIDER IN CLOUD | Science Publications

摘要

> Software as a Service (SaaS) offers reliable access to software applications to the end users over the Internet without direct investment in infrastructure and software. SaaS providers utilize resources of internal data centres or rent resources from a public Infrastructure as a Service (IaaS) provider in order to serve their customers. Internal hosting can ample cost of administration and maintenance whereas hiring from an IaaS provider can impact the service quality due to its variable performance. To surmount these drawbacks, we propose pioneering admission control and scheduling algorithms for SaaS providers to effectively utilize public Cloud resources to maximize profit by minimizing cost and improving customer satisfaction level. There is a drawback in this method is strength of the algorithms by handling errors in dynamic scenario of cloud environment, also there is a need of machine learning method to predict the strategies and produce the according resources. The admission control provided by trust model that is based on SLA uses different strategies to decide upon accepting user requests so that there is minimal performance impact, avoiding SLA penalties that are giving higher profit. Machine learning method aims at building a distributed system for cloud resource monitoring and prediction that includes learning-based methodologies for modelling and optimization of resource prediction models. The learning methods are Artificial Neural Network (ANN) and Support Vector Machine (SVM) are two typical machine learning strategies in the category of regression computation. These two methods can be employed for modelling resource state prediction. In addition, we conduct a widespread evaluation study to analyze which solution matches best in which scenario to maximize SaaS provider?s profit. Results obtained through our extensive simulation shows that our proposed algorithms provide significant improvement (up to 40% cost saving) over literature reference ones.