Energy efficient scheduling of real time tasks on large systems and cloud
No Thumbnail Available
Date
2018
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Large systems and cloud computing paradigm have emerged as a promising computing platform of recent time. These systems attracted users from various domains, and their applications are getting deployed for several benefits, such as reliability, scalability, elasticity, pay-as-you-go pricing model, etc. These applications are often of real-time nature and require a significant amount of computing resources. With the usages of the computing resources, the energy consumption also increases, and the high energy consumption of the large systems has become a serious concern. A reduction in the energy consumption for the large systems yields not only monetary benefits to the service providers, but also yields performance and environmental benefits as a whole. Hence, designing energy-efficient scheduling strategies for the real-time applications on the large systems becomes essential. The first work of the thesis devises a coarse-grained thread-based power consumption model which exploits the power consumption pattern of the recent multi-threaded processors and then proposes three energy-efficient scheduling policies. Experimental results show significant improvement compared to the baselines. The second work of the thesis considers a utilization-based power consumption model for a heterogeneous virtualized cloud system where the utilization of a host can be divided in three ranges and then proposes scheduling technique based on that. The proposed scheduling technique reduces energy consumption by almost 24% w.r.t. the state-of-art policy. As the cloud providers often offer VMs with discrete compute capacities and sizes, which leads to discrete host utilization, the third work of the thesis considers scheduling a set of real-time tasks on a virtualized cloud system which offers Vms with discrete compute capacities. The problem is divided into four sub-problems based on the characteristics of the tasks and solutions are proposed for each sub-problem. The fourth work of the thesis considers scheduling of online scientific workflows on the virtualized cloud system where a scientific workflow is taken as a chain of multi-VM tasks. series of scheduling approaches are proposed considering several options and restrictions on migration, allocation of Vms, and slack distribution. Experimental results show that the proposed scheduling policy under nonsplittable VM allocation category consumes a similar amount of energy as the baseline policy but with a much lesser number of migrations. For the splittable VM allocation category, the proposed policies achieve energy reduction of almost 60% as compared to the state-of-art policy.
Description
Supervisors: Aryabartta Sahu and Sushanta Karmakar
Keywords
COMPUTER SCIENCE AND ENGINEERING