On the Efficiency of Network Architecture and Network Provisioning in Cloud Data Centers

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


PhD Thesis Defence


Title: "On the Efficiency of Network Architecture and Network Provisioning 
in Cloud Data Centers"

By

Mr. Ting WANG


Abstract

Large-scale virtualized data centers enable the new era of cloud computing 
and provide the core infrastructure to meet the computing and storage 
requirements for both enterprise information technology needs and 
cloud-based services. To support the ever-growing cloud computing needs, 
the number of physical servers and virtual machines in today's data 
centers are increasing exponentially, which in turn leads to enormous 
challenges in designing cost-effective data center networks (DCNs) and 
efficient resource sharing via network provisioning. With data 
availability and security at stake, the issues with data center networks 
are more critical than ever. As one of the most important determinants of 
network performance, the data center network architecture plays a dominant 
role in determining the system reliability, network capacity, latency and 
even the routing efficiency. With this motivation, in this thesis paper, 
we firstly present the design, implementation and evaluation of SprintNet, 
a novel server-centric network architecture for data centers. SprintNet 
achieves high performance in network capacity, fault tolerance, and 
network latency. SprintNet is also a scalable, yet low-diameter network 
architecture where the longest path length of SprintNet can be limited by 
four hops and is independent of the number of layers. Subsequently, 
another agile and cost-effective server-centric architecture, termed 
JieLin, is proposed. JieLin not only is highlighted by its excellent 
scalability of O(n*(an)(k-1)), but also demonstrates high performance in 
bisection bandwidth, fault tolerance and average path length. Furthermore, 
in response to the critical shortcomings of the server-centric 
architectures, we then propose an effective hardware based approach to 
solve these issues and improve the network efficiency. Afterwards, we 
design two torus-based DCN architectures named NovaCube and CLOT, where 
the torus topology well implements the network locality forming the 
servers in close proximity of each other, which increases the 
communication efficiency. Moreover, in the highly multiplexed shared cloud 
data centers, in order to efficiently share the physical network resources 
among multiple tenants that have diversified virtual network topologies 
with different network characteristics, in this thesis we then propose an 
efficient online heuristic virtual network embedding framework called 
Presto, adopting an artificial intelligence resource abstraction model 
named Blocking Island. Presto operates with quite low computation 
complexity and greatly reduces the search space, which far outperforms 
other candidates. The goal of Presto is to maximize the economic revenue 
of infrastructure providers and increase the resource utilization while 
minimizing the embedding cost. Finally, we propose two efficient heuristic 
schemes to achieve an energy proportional data center network from the 
perspective of resource allocation, routing and flow scheduling, without 
compromising throughput and fault tolerance too much. Both theoretical 
analysis and extensive simulations have been conducted to evaluate the 
overall performance and effectiveness of these proposals.


Date:			Friday, 28 August 2015

Time:			10:00am - 12:00noon

Venue:			Room 2132B
 			Lift 19

Chairman:		Prof. Ross Murch (ECE)

Committee Members:	Prof. Mounir Hamdi (Supervisor)
 			Prof. Gary Chan
 			Prof. Andrew Horner
 			Prof. Danny Tsang (ECE)
 			Prof. Hussein Mouftah (Inf. Tech. & Engg.,
 					       U of Ottawa)


**** ALL are Welcome ****