Full Program »
A Hybrid Automatic Elasticity Solution for the IaaS Layer based on Dynamic Thresholds and Time Series
Currently, the most popular auto-elastic solutions at cloud providers are rule-based solutions with fixed thresholds. However, these solutions suffer from two main deficiencies: reactive nature and configuration difficulty. In this paper, we propose a hybrid auto-scaling approach that overcomes the deficiencies of the solutions based on rules with fixed thresholds. The propose auto-scaling consists of a hybrid approach, composed of a set of self-adaptive predictions and a reactive component based on dynamic thresholds. The forecast set corrects the first flaw (that is, the reactive nature) of rule-based solutions by predicting the future workload close to the cloud service. Also, the reactive component based on dynamic thresholds reduces the difficulty of configuring the auto-scaling solution. The evaluation results show that the proposed solution reduces average response time for the Wikibench benchmark by up to 70% compared to Amazon’s auto-scaling system.