The number of Data Centers and the servers present in them has been on the rise over the last decade with the advent of cloud computing, social networking, Big data analytics etc. This has eventually led to the increase in the power consumption of the Data Center due to the power hungry interconnection fabric which consists of switches and routers. The scalability of the data center has also become a problem due to the interconnect cabling complexity which is also responsible for the increase in the energy used for cooling the data center as these bundles of wires reduce the air flow in the data center. The maintenance costs of the data center is high due to this reason. This brings the challenge of reducing the power consumption as well as improving the scalability of the data center.
There is a lot of cost involved in the establishment of a network in a data center and this network is one of the main source of power consumption. Therefore, there is a need to accurately characterize the data center network before its construction which requires the simulation of the data center models. For the simulation of data center models, we require the traffic which is identical to that of an actual data center so that the results will be similar to a real time data center.
Traditional data center networks have a wired communication fabric, which is not scalable and contributes largely to the power consumption. This has led to the investigation of other methods. There have been transceivers designed that can support the unlicensed 60 GHz spectrum, supporting high bandwidth similar to the wired network present in traditional data centers. These wireless links have spatial reusability and the data centers can make use of this communication medium to meet the high bandwidth demands and also reduce the use of cable thereby bringing down the cost and the power consumption.
This thesis studies the previous traffic models used in the simulation of a data center network. Traffic collected from ten different data centers is then characterized and modelled based on various probability distributions. The implementation of the model tries to generate traffic similar to that of an actual data center. The Data Center Network is then simulated using the traffic generated and the performance of the wired data center is quantified in terms of metrics like throughput, latency and the power consumption of the data center networks.
Library of Congress Subject Headings
Data libraries--Energy consumption; Cloud computing; Client/server computing
Computer Engineering (MS)
Department, Program, or Center
Computer Engineering (KGCOE)
Chandrasekaran, Siva Shankar, "Understanding Traffic Characteristics in a Server to Server Data Center Network" (2017). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus