An Intelligent Framework for Flavor Recommendation and Cost Optimization in Hybrid Cloud Autoscaling

Agnes Osagie^1,*, Sandra Terazic², Barbara Charchekhandra³

¹Cape Peninsula University of Technology, Faculty of Applied Science, South Africa

²Department of Mathematics, University of Rijeka, City of Rijeka, Croatia

³Jadavpur University, Department Of Mathematics, Kolkata, India

Emails: Osagieagne2000@cput.ac.za; Sandy1997te@Uniri.hr; Charchekhandrabar32@yahoo.com

Abstract

This research presents a flavor recommendation framework that intends to be used in hybrid clouds to address resource provisioning and cost issues. A cloud “flavor” is an instance type that assigns values for CPU, memory, storage, and networking. Today the flavor selection process is manual, and no dynamic technique is used, therefore, the process is inefficient because some flavors are underutilized. The proposed framework also provides flavor recommendations for autopiloted dynamic capacity provisioning using predictor analysis of workload and cost proportional to different CSPs. It uses an RNN_LSTM-based Proactive Predictive Engine (PPE) to quantitatively estimate future resource requirements and a Recommendation Engine consisting of the scoring and flavor engines. This framework receives the application’s actual and predicted consumption of CPU and memory, cost fluctuations, and CSPs’ options and then the selection of various flavors is performed in the runtime. Metrics are gathered, stored, and analyzed in real-time through Telegraf, InfluxDB, and Apache Libcloud for current resource allocation. Experimental results of the system on AWS and OpenStack show the benefit of using the proposed framework, which reduced the number of EBS and VMs by 19% and the cost saving by up to 17% compared with traditional and reactive approaches. This solution turns static resource allocation into a real-time predictive accuracy of how resources are best utilized as well as the expense at the hybrid cloud environment.

Keywords: Autoscaling; Flavor Recommendation; Hybrid Cloud; RNN_LSTM; RPE; CSPs; VMs; Cost optimization; Resource forecasting; CPU utilization; RAM utilization; InfluxDB, Telegraf; Apache Libcloud; Dynamic resource allocation