Cloud Resource Prediction for Healthcare as a Scientific Application

Abstract

Scientific applications demand high performance computing in reasonable time which can be satisfied by using Cloud capabilities. The Scientists can scale up or scale down the infrastructure as per their requirement. Moreover, they need not plan the resources beforehand and can use them on the fly. They can customize the application environment as per requirements. Hence, Cloud is a cost effective, flexible and efficient solution for Scientific applications. Cloud Computing serves multiple consumers on demand by pooling the resources and charge them as per usage. This makes managing the resources easier. Gradually, advancements were done for resource management by incorporating efficient resource scheduling and prediction techniques. Resource prediction helps in proper resource management not only by attaining QoS requirements for the end users but also by satisfying provider by cutting costs of resources. A lot of research work has been done for Cloud resource prediction, but the existing techniques are not specific to Scientific applications. Moreover, these techniques either satisfy end user or provider. This thesis explores the gaps in the existing techniques and analyses them to propose a Resource Prediction Model in Cloud environment. The model analyses the application details, monitors the CPU Utilization and execution time, evaluates the optimal number of virtual machines for maximizing CPU Utilization and minimizing execution time, trains a model using the collected dataset and finally perform predictions for a similar application. In order to implement the Resource Prediction Model, a data mining healthcare application, named PharmacoGen was developed and proposed as a case study. The different phases of the proposed model were implemented using the case study to validate the model. The predicted VMs were validated to be having maximum ratio of CPU Utilization and execution time by simulating the application in CloudSim. The proposed model aims to eliminate the research gaps by providing the optimal number of resources for any kind of Scientific application by considering its parameters. It satisfies both the end user and provider as it selects VMs which maximize the CPU utilization and minimize the execution time.