Date of Award
12-2018
Document Type
Thesis
Degree Name
Master of Science (MS)
Department
Electrical and Computer Engineering (Holcomb Dept. of)
Committee Member
Dr. Kuang-Ching Wang, Committee Chair
Committee Member
Dr. Harlan Russell
Committee Member
Dr. Walt Ligon
Abstract
Modern cloud applications are hosted on data centers across vast geographical scopes and exchange large amounts of data continuously. Transmission Control Protocol (TCP) is the most popular protocol for reliable data transfer; however, due to TCP’s congestion control mechanism, maximum achievable throughput across a large bandwidth-delay product (BDP) network is limited. Various solutions exist to enhance data transfer throughput but they usually require non-trivial and explicit installation and tuning of specialized software on both sides which makes deployment limited. A software defined networking (SDN) based solution "Steroid OpenFlow Service (SOS)" was developed that utilizes multiple parallel TCP connections to transparently enhance network performance across a large BDP network. OpenFlow is used to transparently redirect user traffic to nearby service machines called SOS agent and these agents use multiple TCP connections to transfer data fast across large BDP network. While SOS has shown significant improvements in data transfer throughput, there are multiple factors which affect its performance. This study focuses on SOS scalability analysis targeting four critical factors: CPU utilization of SOS agents, sockets used for parallel TCP connections, how OpenFlow is used and network configurations. Through this study, the SOS agent code was revamped for performance improvements. Experiments were conducted on the National Science Foundation’s CloudLab platform to assess the effect of the above-mentioned factors on SOS performance. Results have shown improvement in throughput per SOS session from 10.96Gbps to 12.82Gbps by removing CPU bottleneck on 25Gbps network. SOS deployment over an InfiniBand network has shown a linear increase in throughput to 23.22Gbps with optimal network configurations. Using OpenFlow to support multiple client connections to the same server have increased throughput from 12.17Gbps to 17.20Gbps. The study showed that with code-level improvements and optimal network configurations, SOS performance can be improved substantially.
Recommended Citation
Anjam, Khayam, "Steroid OpenFlow Service Scalability Analysis" (2018). All Theses. 2965.
https://open.clemson.edu/all_theses/2965