Design and Evaluation of Data Center Network Topologies

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering


PhD Thesis Defence


Title: "Design and Evaluation of Data Center Network Topologies"

By

Mr. Yang LIU


Abstract

Large-scale data centers form the core infrastructure support for the ever 
expanding cloud based services, thus the performance and dependability 
characteristics of data centers will have significant impact on the 
scalability of these services. In particular, the data center network 
(DCN) needs to be agile and reconfigurable in order to respond quickly to 
ever changing application demands and service requirements. Significant 
research work has been done on designing the data center network 
topologies in order to improve the performance of data centers.

In this thesis, we first present a abstract survey of DCN designs and 
topologies that have published recently. We start with a discussion on 
various representative data center network topologies, and compare them 
with respect to several properties in order to highlight their advantages 
and disadvantages. A good understanding of the state-of-the-art in DCN 
would enable the design of future architectures in order to improve 
performance and dependability of data centers.

With the plethora of DCN topologies that have been proposed in the 
literature, there is a sore need for a standardized method for evaluating 
and comparing various alternate DCN architectures. Considering these 
needs, we designed DCNSim, a general purpose DCN simulator that supports 
most well-known DCN topologies proposed in the literature. Our simulator 
can generate various metrics for the topologies. The modular and flexible 
architecture of the simulator permits easy extension to support any future 
proposed topologies and compute new metrics.

With the support of DCNSim, we present evaluations of the fault-tolerance 
characteristics of several important DCN topologies. These enable us to 
present an objective comparison of the topologies under faulty conditions. 
Besides individual failures, we introduce the idea of fault regions to 
study associated failures in DCNs. Evaluations regarding fault regions 
give us a more comprehensive view on the performance of DCN topologies.

Optical DCNs are more complex than the architectures discussed above 
because they do not have fixed topology. However, they are becoming 
increasingly popular for their technological advantages including 
flexibility and high bandwidth. The existing algorithm applied in 
literature cannot fully utilize such model and its flexibility. Besides, 
reconfiguring optical DCN topology according to the changing traffic is a 
problem that has never been well researched. We focus on addressing these 
two main questions which are of key importance to fully exploit the 
advantages of the optical DCNs. To this end, we present algorithms to 
solve the problems.


Date:			Friday, 23 August 2013

Time:			2:00pm - 4:00pm

Venue:			Room 3501
 			Lifts 25/26

Chairman:		Prof. Qingping Sun (MECH)

Committee Members:	Prof. Jogesh Muppala (Supervisor)
 			Prof. Gary Chan
 			Prof. Kai Chen
 			Prof. Amine Bermak (ECE)
                        Prof. Malathi Veeraraghavan (Elec., & Comp., Engg.,
 						     Univ. of Virginia)


**** ALL are Welcome ****