Virtual Resource Provisioning In Clouds: A Cost Saving Approach

This paper addresses the problem of cost optimized resource provisioning for cloud applications. With the fast proliferation of cloud industry, the cost-effectiveness of cloud computing is becoming more important than ever. To achieve cost-effectiveness for an application provider, an efficient method is cost optimization, which inevitably leads to reducing resources. However, a possible result of reducing resources is the decline of application performance, which further leads to SLA violation and punish-ment. On the other hand, existing resource pricing could be used to optimize cost even though the total resource amount is fixed. Hence, we designed a framework to make predictions and provision resources accordingly to achieve the goal of cost optimization for application providers.For workload prediction, we investigated some commonly used methods and mod-els, and found that different workload scenarios may suit to different models. Thus we designed an interface for workload prediction and implemented a default ARIMA model, which makes predictions to help provision resources. In this way, a default prediction model is available to users and it is also easy to replace for further improve-ments.Then we designed the workload-resource projection module to figure out the map-ping relationship between workload and VM amount. To achieve cost optimization, we must consider the relationship between workload, resources and application per-formance. Usually, to a fixed workload, resources and application performance are positively related; while to a fixed amount of resources, workload and application per-formance are negatively related. Hence resources should be provisioned according to workload in order to avoid resource waste to achieve cost optimization as well as to ensure application performance. The workload-resource projection module provisions resources according to predicted workload and ensures application performance.After the VM amount is resolved, by exploiting pricing policies of existing cloud providers, our framework presents a hybrid resource provisioning solution, which in-volves instances of different prices and leasing durations. It furthers cost optimization for application providers.We used the 1998 WorldCup workload dataset to verify the effectiveness of our framework on CloudSim platform. The prediction results show that the prediction model we implemented has a comparatively low relative error. And the workload-resource projection module ensures the QoS requirement in most time periods. The resource provisioning module resolves a Hybrid resource provisioning solution, which saves a considerable part of cost compared to other resource solutions. Thus our frame-work achieves the cost optimization goal without doubt.