Investigating the Performance of Machine Learning Algorithms for Improving Fault Tolerance for Large Scale Workflow Applications in Cloud Computing

摘要

Cloud platform is emerging distributed system mainly used to host applications from the client side. Maintenance of user’s data and execution of client’s application with the help of hardware, software and network resources are allowed in cloud distributed environment. In this era, modern cloud data centers are used for hosting many non-commercial applications used in scientific field. Cloud being a distributed platform, many errors and fault can occur. In such a distributed environment it becomes difficult to identify errors and faults. Hence there is a high requirement for implementation of fault tolerance mechanism in cloud platform. This mechanism ensures that although the failure occurs in cloud the client’s data are not affected in any manner. Cloud’s performance can be improved by ensuring users their on-demand services as required with the help of fault tolerance mechanisms. In this research work the comparative analysis of three machine learning algorithms K-Means, Decision Tree and KNN (K Nearest Neighbors) algorithm for different structures of scientific workflow applications such as Pipeline, Merge, Split, Diamond using parameters like Sensitivity and Specificity are discussed.