Energy-Efficient Resource Scheduling Techniques For Cloud Computing

Abstract

Cloud computing, a form of distributed computing, assures steadfast and convoluted service delivery. The emergence of cloud has revolutionized the style of consumption and delivery of computing and information technology by triggering relocation of computing and data capabilities from personalized computers to big data centers. The setting up of cloud data centers has embellished the delivery of extensive, reliable and intricate technical developments while assuring extremely networked, scalable, and virtualized environment to the users. The data centers facilitating cloud services have undoubtedly ensured swift service delivery but have signi cantly exaggerated the energy demands and caused severe energy crisis. A large amount of energy is wasted in these data centers whilst posing performance abasements due to irregular resource utility and energy consumption levels. The trepidations related to energy crisis have further rooted serious environmental concerns and danger to ecological sustainability by increasing atmospheric Green House Gas and Carbon Dioxide (CO2) emissions. Besides this, the high energy consumption and demand hampers processing capabilities in terms of execution time, Quality of Service (QoS) parameters etc. and elevates energy-related expenditures. Consequently, the realization of energy e ciency and surmountability of high energy demands has become prime concern for the computing sector. The curtailment of the energy demands becomes more important when scarcity, location-dependency and other restrictions are associated with the renewable and non-renewable sources of energy. Thus, the constraints related to such explicit measures and mounting energy demands have called for the implicit management of the energy crisis within the data centers. For this, energy optimization initiatives have been made at both the hardware as well as software levels, out of which, the software-oriented measures have proved more signi cant in terms of accomplishment and cost. The energy-aware allocation and scheduling schemes form part of such software-based energy e ciency measures.The research work presented in this thesis proposes energy-e cient resource scheduling techniques for cloud. The techniques are o ered through two energye cient allocation and scheduling models in cloud computing. These models o er e cient resource utilizations, pro cient energy usages, improvement in the performance while minimizing energy costs. The rst model is Green Cloud Scheduling Model (GCSM) that performs energy-e cient allocation and scheduling of tasks on the cloud nodes. It principally tends to optimize the energy consumption and thus referred to as green model. The Green Cloud Scheduling (GCS) technique of GCSM tends to optimally provision and schedule the tasks entering the system while maintaining minimal energy consumption, maximizing utilization of resources, ful lling task deadlines and averting degradation in the overall performance. Exclusively, GCSM is an energy and deadline optimization model in cloud computing. Thereafter, a GreenSched model, an extended model of GCSM has been proposed. GreenSched implements an intelligent machine learning technique, Forwardonly Counter Propagation Network (CPN)-based scheduling (FCS), that proactively allocates and schedules deadline-and-budget constrained tasks to identi ed energyaware cloud nodes. GreenSched primarily achieves energy e ciency along with optimizing variable QoS metrics such as, deadline, budget and performance. Contrary to GCSM, GreenSched model is an intelligent model that apart from o ering energy e ciency also assures high QoS for performance and cost parameters. Unlike GCSM, it assures contentment of both user-imposed deadline and budget restrictions particularly when energy, deadline and budget are closely inter-related parameters. The performance of the proposed models with variation in time, frequency of tasks and number of nodes has been analyzed which vary in the cloud environment. Also, all the proposed techniques have been evaluated with respect to the energy consumption levels and performance. In order to validate GCSM, a prototype environment as close as possible to a cloud data centre has been simulated. A heterogeneous cloud environment consisting of di erent virtualized servers to exploit the power of cloud computing has been created. The experimental results predict that GCSM achieves energy savings up to 71% and almost 82% tasks are completed within the deadline constraints as imposed by users pertaining to each task. For evaluation of GreenSched model, a simulated data center in CloudSim toolkit has been created consisting of variable nodes, running multiple VMs using the customizable policies and methods available in the CloudSim toolkit. The experimental observations indicate that the scheduling technique in GreenSched model achieves an overall energy savings up to 64.21% as compared to two other techniques. Thereafter, the proposed GCS scheduling technique of GCSM has been experimentally verified through a case study conducted at Corporation Bank, Bakkarwala branch, New Delhi. GCS has been experimentally compared for energy consumption levels, resource utilizations (both CPU and memory) and performance achieved with cloud-based Finacle application scheduler implemented by the Corporation Bank. The experimental results demonstrate the e fficacy of GCS by saving an average of 76.58% energy with 80% CPU and 67% memory utility levels. GCS achieves 81.3% performance while minimizing the energy and thus achieving green banking operation at the bank. The testing results demonstrate that the proposed solutions are working e fficiently and can be effectively used to address the low resource utility and high energy consumption challenges to realize energy e fficiency. By minimizing the energy consumption, the proposed models and their techniques circuitously reduce energy demands, carbon emissions and tend to overcome cooling requirements, energy expenditures of the cloud data centers and help in accomplishing green computing.