Ahmad Ghalayini, Sina Jandaghi Semnani
Original Paper: Singla, Ankit, Philip Brighten Godfrey, and Alexandra Kolla. “High Throughput Data Center Topology Design.” NSDI. Vol. 14. 2014.
The paper we are reproducing tries to answer several interesting questions concerning data center topology design, especially the ones related to random graph topologies. It first demonstrates how random graphs in a homogeneous data center topology design achieve a throughput close to the theoretical upper bound derived in the paper. Then, more interestingly, the authors conduct a systematic search in the space of random graphs to find the best placement, clustering, and interconnection of servers and switches in heterogeneous1 networks. They finally apply what they deduced from their experiments and simulations to improve VL2 topology for heterogeneous networks, resulting in 43% higher throughput with the same equipment. For all sections of the paper except the last one (Section 8), the authors do not consider systems-level issues such as routing and congestion control; instead, they model network traffic using fluid splittable flows which are routed optimally. In this project, we replicate their work by plotting the most important theoretical figures in the paper. For this purpose, we had to infer some of their hidden assumptions, and learn how to use the tools they are using. At the end of the project, we were able to reproduce the figures with a very good similarity.