Date of Award
9-4-2015
Document Type
Thesis
Degree Name
Computer Science, MS
First Advisor
Hai Jiang
Committee Members
E Hammerand; Xiuzhen Huang
Call Number
LD 251 .A566t 2015 G64
Abstract
MapReduce is a programming model that is widely espoused in the industry and viewed as one of the best options for data-intensive operations. Map and Reduce functions are two core building blocks of the model. MapReduce libraries execute these functions in parallel and distributed mode on massive clusters to achieve high throughput. This thesis extends the same methodology by leveraging Graphic Processing Unit (GPU) powered nodes in the cluster to enhance the performance to next level. The system works at three different levels i.e. Cluster, CPU and GPU and aims to exploit all the available resources. The main focus of the thesis is on the third tier (GPU) because of the challenges associated and scope of performance increase when those challenges are addressed. The system executes the user-defined Map and Reduce functions on the input data by distributing the job among available working elements at every layer in an iterative manner and then by collecting the processed results in the same manner but in reverse order. Experiments show that the performance increased substantially with the introduction of each tier in the system. The outcome of the research can be combined with other studies done in the area to further refine the application performance.
Rights Management
This work is licensed under a Creative Commons Attribution 4.0 International License.
Recommended Citation
Gupta, Harshit, "Three Tier MapReduce" (2015). Student Theses and Dissertations. 730.
https://arch.astate.edu/all-etd/730
