Off-campus WSU users: To download campus access dissertations, please use the following link to log into our proxy server with your WSU access ID and password, then click the "Off-campus Download" button below.
Non-WSU users: Please talk to your librarian about requesting this thesis through interlibrary loan.
Date of Award
In recent history, there has been a rapid growth in the amount of data created across the globe. This data has been created in variety of formats due to growing sense of usefulness of data in numerous industries like Automotive, Social Media, Retail, E-commerce, Healthcare,
etc. Collectively simple activities performed by individuals or machines in this world create variety of high volume data at a fast pace. Predictions have been made that the total data volume in the digital world will grow 1000 times more than the volume of data in present in year 2016. The extra-ordinary growth in three V’s (Volume, Variety and Velocity) creates new challenges to handle Big Data. This increasing demand to efficiently handle ever growing data gave birth to distributed data storage and processing frameworks such as Hadoop, Spark, Mahout, Storm, Flink, etc. However, with increased number of available options to process the big data, comes a new challenge to select a perfect tool according to requirements. Selecting an optimal tool to perform some specific operations on specific type of data requires great
deal of efforts in researching the working principals of the tool. In this study, we provide detailed information about underlying processing methodology, component stack used by both Spark and Flink along with key differences in them. We perform extensive set of experiments
on multiple problems including batch jobs, iterative and machine learning jobs and streaming jobs. Finally, we provide detailed analysis of the performance comparison and the reasoning behind it.
Jadhav, Niranjan, "Comparing Performance Of Spark And Flink On Batch And Streaming Data" (2017). Wayne State University Theses. 621.