Proactive Fault Tolerance Technique for Scientific Applications in Cloud

Abstract

Cloud Computing means hosted services are delivered over Internet. Instead of building their own computing infrastructures, consumers can buy the computing resources like virtual machines, storage as a utility just like an electricity. Cloud computing gives advantages to consumers like on demand service, rapid elasticity, resource sharing and pooling, scalability and many more. But there are many issues related with the cloud computing which need to be addressed properly like fault tolerance, resource scheduling and effective pricing of cloud resources etc. Making cloud environment fault tolerant is the major challenge these days. In order to improve reliability and achieve robustness in cloud computing, failures should be assessed and handled effectively. As the cloud environment is dynamic, unpredictable system operation occurs, causing problems and failures. When Cloudlets or tasks of any application are executing on Cloud platform , then effective allocation of cloud resources like virtual machines to the cloudlets is very important such that no cloudlet gets failed or no virtual machine is left out for the execution. If there is no proper allocation of cloud resources to the cloudlets , it is considered as a big failure in the effective working of any application on Cloud platform. This fault or failure needs to be handled carefully by the fault tolerance techniques. The used techniques for the fault tolerance comes in the category of reactive (fault is handled after it has occurred) and proactive (fault is predicted before it actually happens).Much work has been done in the area of reactive techniques. The existing proactive techniques are not that much reliable in fault tolerance as compared to reactive techniques. Moreover, fault tolerance techniques work more efficiently in case of real time applications rather than scientific applications .Consequently, an approach based on proactive technique need to be proposed that can proficiently resolve the fault tolerance problem for effective allocation of cloud resources for scientific applications. This thesis focuses on providing an efficient proactive fault tolerant technique for scientific application in which above mentioned fault is predicted proactively and handled efficiently such that all cloudlets finish their tasks before or on deadline and no virtual machine has been left out for execution.