Data center ethernet : data traffic congestion control

• Main Author: Chua, Kian Siong.
• Other Authors: Foh Chuan Heng
• Format: Final Year Project
• Language: English
• Published: 2012
• Subjects: DRNTU::Engineering::Computer science and engineering::Computer systems organization::Computer-communication networks; DRNTU::Engineering::Computer science and engineering::Data::Data storage representations
• Online Access: http://hdl.handle.net/10356/48524
• Institution: Nanyang Technological University

With advancements in cloud technology, more and more applications are turning towards the cloud to host their data. These applications are diverse and have different requirements including low latency for short flows to needing high throughput and utilization for long flows. The transport control protocol (TCP) might serve to be good enough for general usage, but specialized infrastructure such as data centers require more specialized protocols. The Data Center Transport Control Protocol (DCTCP) is a modification of TCP. It makes use of a previously proposed Explicit Congestion Notification (ECN) system, which is also an improvement from TCP. In TCP, hosts determine the ideal congestion window size by increasing it until there is a packet loss. Incorporating the ECN system however, allows the hosts to reduce their congestion window before a packet loss occurs. This is possible because the ECN system reacts to the extent of congestion and takes measures to control it before a packet loss occurs. In this way, packet losses are reduced and higher throughput can be achieved. DCTCP was designed to suit data centers which have three basic requirements. Firstly, it must have low latency for short flows. Secondly, a high burst tolerance. And lastly, high utilization or throughput for long flows. The DCTCP algorithm claims to achieve these three requirements. In this report, we put these claims to test to see if the DCTCP algorithm does actually meet these requirements, and if they do not, we seek to provide answers to why our results differ from the ideal and how to best improve on it.