Performance Analysis of Realtime Task Scheduling in Cloud System

Abstract

Cloud computing is becoming an important computing paradigm due to its costefficiency, scalability, availability, and high resource utilization. Applications like financial transactions, healthcare, scientific workflows, video streaming, Internet of Things (IoT), etc. with their realtime nature, need provisioning of cloud resources to guarantee timeliness and high availability. The Cloud Service Providers (CSPs) must support sufficient cloud resources to satisfy the demand from these realtime applications. Meanwhile, the evergrowing demand from applications forces CSPs to deploy more and more cloud resources, which consumes a considerable amount of energy. The high energy consumption affects the environment and other metrics like execution cost, makespan, and reliability of the cloud system. Hence, it necessitates employing some techniques to reduce cloud systems’ energy consumption and make it energyefficient along with other performance metrics like reliability, execution cost, makespan, etc. Realtime task scheduling is one of the methods to achieve energyefficiency in the cloud system. Moreover, heterogeneous computing environments and application timing constraints add complexity to the realtime task scheduling. Therefore, the study of a cloud system’s performance is necessary for realtime applications to ensure Quality of Service (QoS), defined in terms of energy consumption, makespan, execution cost, reliability, etc.
First, an Energy and Cost Aware task scheduling (ECA) algorithm based on the TOPSIS analysis method is proposed to reduce energy consumption and execution cost. Here, a scoring value is calculated for a VM based on its energy consumption and execution cost to execute a task. Then, a VM with the best score in terms of energy usage and execution cost is selected. Next, a Learning Automata (LA)based scheduling (LAS) algorithm is proposed to minimize energy consumption and makespan. It is a reinforcementbased method where the action, i.e., assignment of a task to a VM, is penalized if it contributes to scheduling objective degradation and rewarded if the action is suitable to improve scheduling objective.The above process is continued for a fixed number of iterations, and the actions with the best reward value are added in the scheduling decision.
Then, a game theory based scheduling algorithm is proposed to enhance system performance where energy consumption and reliability are considered as the performance metrics. The biobjective scheduling algorithm is modeled as a noncooperative scheduling game, named Realtime Task Scheduling Game (RTSG). The solution or Nash Equilibrium of RTSG is presented using a Vickery auction mechanism. The proposed solution is compared with a cooperative game based solution and an auctionbased approach.
Finally, a faulttolerant scheduling algorithm is presented, taking into account energy consumption and reliability. First, an acceptance test mechanism is designed considering schedulability and response time failure to detect VM failure. A reliability and energyaware Faulttolerant scheduling algorithm, REO, is proposed using the PB concept and BB overlapping technique. The performance metrics used for comparison of algorithms include Success Ratio, makespan, and total energy consumption. The outcomes of the simulation results signify the usefulness and effectiveness of the proposed algorithms for studying realtime task scheduling performance.