Improving Research and Education of Big Data and Cloud Computing at Winston-Salem State University

Winston-Salem State University, Winston Salem NC

Investigators

Abstract

The Historically Black Colleges and Universities - Undergraduate Program provides support for projects that offer solutions to the severe underrepresentation of African American students in computer science at the undergraduate level. The project at Winston Salem State University will build research capacity among faculty and students in the Department of Computer Science by examining resource management of big data applications on cloud infrastructure. This work will be done in collaboration with Duke University. Undergraduate computer science majors will gain research experiences that may further their interest in pursuing a degree in the discipline, thereby contributing to broadening the participation of underrepresented groups in the discipline. The project goals are: to build the big data and cloud computing research capacity at Winston-Salem state university; to involve undergraduate students in cutting-edge scientific research; and to enhance educational experiences for computer science majors. The project is exploring research centered around the estimation of a globally efficient and cost effective resource allocation for concurrent MapReduce workloads in a cloud infrastructure so that the configuration ensures meeting service level objectives (SLOs) of various workloads and at the same time minimizing the resources required to achieve this goal. The specific research aims are: to leverage a profile and prediction subsystem in order to characterize and model the performance of a MapReduce workload and identify them accordingly as small jobs, periodic jobs, and delayed jobs; to investigate the parameters to be included in the utility function that best can capture the priority, deadline, performance characterization and heterogeneity of MapReduce workloads and the heterogeneity and dynamics of underlying cloud infrastructure; to design and develop algorithms and heuristics for utility-based resource management that take advantage of the workload-specific characteristics and deploy them with the goal of meeting SLOs and maximizing cloud utilization; and to evaluate the proposed approach with real-life big data workloads and study their performances, cloud rental costs, and the overall utilization of the cloud resources.