QoS-aware Autonomic Resource Provisioning and Scheduling for Cloud Computing

Abstract

Cloud computing is a new paradigm that provides on-demand services over the Internet. Cloud services are viewed as a composition of distributed components and are offered as: Infrastructure (hardware, storage, and network), Platform or Software. On-demand, flexibility and availability of computing components are the main success factors of cloud computing. However, providing dedicated cloud services that ensure dynamic user's Quality of Service (QoS) requirements is a big challenge in cloud computing. Currently, cloud services are provisioned according to resources’ availability without ensuring the expected performances. To overcome this, there is a need to consider two important aspects which reflect the complexity introduced by the cloud management: QoSaware and autonomic management of cloud services. QoS-aware aspect involves the capacity of a service to be aware of its behavior to ensure the elasticity, high availability, reliability of service, cost, time etc. Autonomic management implies the fact that the service is able to selfmanage itself as per its environment needs. Thus, integration of QoS-aware aspects in each cloud component in order to control and inform the system about its current behavior is required. As more and more users give their applications to cloud providers, Service Level Agreements (SLAs) between clients and providers appear as a key representative. Due to the dynamic nature of the cloud, endless supervision on QoS attributes is necessary to impose SLAs. The success of next-generation cloud computing infrastructures will depend on how proficiently these infrastructures will be able to instantiate and sustain computing platforms, which meet randomly varying resource and service requirements of cloud costumer applications. Logically, based on QoS requirements such as scalability, high availability, energy, trust and security, these applications will be characterized and identified in SLAs. This work focuses on to develop a resource provisioning and scheduling framework (QUORA) that automatically manages QoS requirements of cloud users and is based on energy efficient usage of cloud infrastructure. To achieve this, a comprehensive investigation has been conducted to study various existing resource provisioning algorithms in cloud computing that is accomplished by in-depth learning of autonomic resource provisioning techniques. Along with that scheduling techniques like bio-inspired, nature inspired and other optimization techniques have been explored for resource scheduling. Initially, a detailed review of the work done in the area of cloud resource management has been done, further existing resource provisioning and scheduling techniques as well as autonomic resource management techniques have been analysed and compared. Based on literature survey, it is apparent that issues of SLA violation, resource contention, provisioning and scheduling are the main challenges besides numerous other issues that need to be addressed. To address these diverse cloud resource management issues, QUORA has been initially proposed and further designed, developed and tested in this research work. The proposed framework caters to provisioned resource distribution and scheduling of resources automatically. QUORA has been divided into three different stages. At first, QoS based cloud resource provisioning technique (Q-aware) has been proposed. The main aim of this technique is to analyze the workloads and then categorize them on the basis of workload patterns. QoS metrics for every QoS requirement of each workload are identified. Further, to find the importance of a quality attribute, weight for every cloud workload is calculated. The workloads are then clustered through K-Means clustering algorithm on the basis of weights assigned and their QoS requirements and are then provisioned before actual scheduling. CloudSim based experimental results demonstrate that Q-aware reduces the execution time up to 16.67% and execution cost up to 28.99% as compared to non-QoS based resource provisioning technique. Secondly, a QoS based resource scheduling technique (QRST) has been designed that takes the provisioned set of resources as input and schedules by efficient utilization of these resources while reducing the SLA violations at runtime and thus achieving cost-effectiveness and desired performance. Resource scheduling is done on the basis of four resource scheduling policies (Compromised Cost - Time based scheduling policy, Time based scheduling policy, Cost based scheduling policy and Bargaining based scheduling policy) and their corresponding algorithms have been proposed based on different scheduling criteria. Thus, execution of cloud workloads to the corresponding resources is done by choosing the appropriate resource scheduling policy. The performance of the proposed policies has been evaluated with existing scheduling policies in CloudSim based simulated cloud environment. Experimental results demonstrate that QRST reduces the execution time by up to 30.94%, energy consumption by up to 17.66% and execution cost by up to 22.72% compared to existing resource scheduling techniques. Finally, a QoS based autonomic resource management technique (CHOPPER) has been proposed which manages resources automatically and provides reliable, secure and cost efficient service by considering four steps (monitor, analyze, plan and execute) of IBM’s autonomic model. Proposed technique considers four properties of self-management: selfhealing, self-configuring, self-optimizing and self-protecting. Tools used for setting up cloud environment for verifying QoS based autonomic resource management technique are Aneka, SNORT, Microsoft Visual Studio 2010, SQL Server 2008, and JADE Platform (for agents). At platform level, Aneka cloud application platform is used as a scalable cloud middleware to make interaction between IaaS and SaaS, and continually monitor the performance of the system. At Infrastructure level, three different servers (consisting of virtual nodes) have been created through Citrix Xen Server and SQL Server has been used for data storage. Experimental results demonstrate that CHOPPER improves the energy efficiency by 9.46%, resource utilization by 18.88%, throughput by 26%, availability by 8.66% and reliability by 9.11% and it reduces the waiting time by 7.35%, SLA violation rate by 7.66, execution time by 12.94% and execution cost by 34.65% as compared to non-autonomic resource management technique. To validate the proposed solution, two case studies have been presented. Firstly, fuzzy logic based energy-aware autonomic resource scheduling technique (EARTH) for cloud has been proposed for energy efficient scheduling of cloud computing resources in data centers. Secondly, cloud based autonomic information system (Agri-Info) for agriculture service has been proposed which manages various types of agriculture related data based on different domains. The performance of both the case studies have been evaluated in real cloud environment and experimental results show that the case studies performs better as compared to existing techniques. The proposed framework has been further validated with existing frameworks like COCCUS and CAS.