Date of Award

4-2016

Degree Type

Thesis

Degree Name

Master of Science (MS)

Department

Computer and Information Technology

First Advisor

Ioannis Papapanagiotou

Second Advisor

Thomas J. Hacker

Committee Chair

Ioannis Papapanagiotou

Committee Co-Chair

Thomas J. Hacker

Committee Member 1

Baijian Yang

Abstract

With the tremendous growth in the amount of information stored on remote locations and cloud systems, many service providers are seeking ways to reduce the amount of redundant information sent across networks by using data de-duplication techniques. Data de-duplication can reduce network traffic without the loss of information, and consequently increase available network bandwidth by reducing redundant traffic. However, due to the heavy computation required for detecting and reducing redundant data transmission, de-duplication itself can become a bottleneck in high capacity links. We completed two parts of work in this research study, Hardware Accelerated Redundancy Elimination in Network Systems (HARENS) and Distributed Redundancy Elimination System Simulation (DRESS). HARENS can significantly improve the performance of redundancy elimination algorithm in a network system by leveraging General Purpose Graphic Processing Unit (GPGPU) techniques as well as other big data optimizations such as the use of a hierarchical multi-threaded pipeline, single machine Map-Reduce, and memory efficiency techniques. Our results indicate that throughput can be increased by a factor of 9 times compared to a naive implementation of the data de-duplication algorithm, providing a net transmission increase of up to 3.0 Gigabits per second (Gbps). DRESS provides further acceleration to the redundancy elimination in network system by deploying HARENS as the server's side redundancy elimination module, and four cooperative distributed byte caches on the clients' side. A client's side distributed byte cache broadcast its cached chunks by sending hash values to other byte caches, so that they can keep a record of all the chunks in the cooperative distributed cache system. When duplications are detected, a client's side byte cache can fetch a chunk directly from either its own cache or peer byte caches rather than server's side redundancy elimination module. Our results indicate that bandwidth savings of the redundancy elimination system with cooperative distributed byte cache can be increased by 12% compared to the one without distributed byte cache, when transferring about 48 Gigabits of data.

Download

Share

COinS