QoS based Scheduling in Cloud Computing

Abstract

Cloud computing is the latest evolution in the distributed computing paradigm and is being widely adopted by enterprises and organizations. The inherent bene ts like instant scalability, pay for use, rapid elasticity, cost e ectiveness, self-manageable service delivery and broader network access makes cloud computing 'the preferred platform' for deploying applications and services. However, the cloud technology being in nascent stage needs to be proven. The biggest challenge confronting cloud service providers is related to e cient provisioning and scheduling of cloud resources. The resources are dynamically pooled in or pooled out as per the application workload representing the elastic model of cloud computing. Henceforth, in such a subscription based cloud computing environment, cloud service providers must perform e cient and cost-e ective scheduling of cloud resources so that cloud users may reap the bene ts of this technology. QoS based resource scheduling is thus a challenging task in a dynamic cloud environment. To achieve the set objectives of addressing cloud resource scheduling challenges, a comprehensive literature review on cloud resource provisioning and scheduling has been done. Prevalent approaches of resource provisioning and scheduling have been explored and a comparative study of cloud schedulers is accomplished to identify their inherent limitations. Further, existing research in cloud computing and its applications has been carried out with special focus on cloud based health care services. Based on the literature survey, it is apparent that biggest challenge confronting cloud service providers is related to QoS based resource scheduling that needs to be addressed. To address the diverse challenges of cloud resource scheduling, a QoS based resource Scheduling Framework (QSF)has been proposed in this work. The proposed framework considers execution of heterogeneous applications comprising of multi-tier web applications corresponding to the transactional workloads and the compute intensive HPC jobs. QSF accomplishes scheduling as a two step process- resource provisioning followed by resource scheduling. To cater to the provisioning needs of transactional applications, QSF takes into account the transient behavior of applications arising due to varying workloads and uncertain performance characteristics of cloud resources. The framework interacts with the user through workload analyzer that fetches applications and user speci c parameters for execution of cloud applications. Further, QoS Mapper component of the framework extracts the information from workload analyzer to perform mapping of application speci c QoS attributes to resource speci c attributes. The mapping operation is accomplished by facilitating scheduler with the behavior analysis information retrieved from behavior analyzer components. The behavior analysis is accomplished at two levels- Application centric and Resource centric. For the application centric behavior analyzer, the system is modeled as a closed form queuing network model. Mean Value Analysis (MVA) algorithm is executed underneath to predict the response time and utilization values with respect to changing workload mix. The operation of resource centric behavior analyzer is based on the observation that any application executing on cloud infrastructure follows a resource usage pattern.For derivation of the patterns, a pattern predictor component of the framework comes into action. It utilizes the state-of-the-art statistical prediction techniques to forecast the future resource usage needs of di erent applications.The execution of behavior analyzers thus enables resource scheduler to estimate the resource requirement of applications for performing scheduling on the provisioned resources. A QoS based Scheduler (QoS-Sched) is accountable for performing the scheduling operations. It creates a pool of resources from the already provisioned machines for scheduling HPC jobs. New user requests are accommodated by proportionally allocating the tasks to the low utilized machines based on user priority and enunciated QoS criteria. The scheduler estimates the attainable values of QoS parameters from the cloud environment and compares with the user speci ed values to compute QoS deviation value. In case a violation in service is expected, a discounted-price policy has been presented to encourage customers to postpone their requests. This eventuates to higher pro ts with minimum QoS violations while retaining the customer base. Thus, the framework successfully accomplishes the resource scheduling operations. The enforcement of the proposed framework has been carried out with a real multitier SaaS application namely 'Cloud Based Intelligent Health Care Service (CBIHCS)'. The said application has been proposed, designed and further deployed on Amazon EC2 cloud environment for providing cloud based health care service solution.CBIHCS advocates the use of cloud technologies for the creation and management of health care services by integrating wireless sensor technologies and mobile computing with cloud based services. In this speci c implementation, CBIHCS performs real time monitoring of user health data for diagnosis of chronic illness such as diabetes. It assimilates user health speci c details and stores in cloud based storage repositories for e cient retrieval and quick updates. Techniques of data mining have been utilized for classi cation of user as Diabetic or Non-Diabetic. The functionalities supported by CBIHCS have further been exploited for creation of multiple test plans and subsequent generation of heterogeneous workloads.In particular,the login session of an anonymous user such as government agency for the purposes of fetching heath records of users for survey purposes corresponds to the transactional workload while the classi cation of users as Diabetic or Non-Diabetic involving complex data mining operations represents compute intensive independent jobs. The workload data has been extracted using JMeter and Amazon Cloud Watch utilities for evaluation of the proposed scheduling policy. For veri cation of the proposed framework, performance evaluation and result comparisons have been accomplished using the CloudSim toolkit. Multiple runs for the simulation experiments have been performed to analyze the e ect on QoS parameters. The QoS based evaluation parameters include resource utilizations, total cost and number of SLA violations along with the computation of number of user requests completed with discounted price policy and without discounted price policy. Finally, the framework has been compared with existing scheduling frameworks to validate the outcomes. The results show that the proposed QoS based resource scheduling framework e caciously addresses the challenges of cloud resource scheduling.