Electrical and Computer Engineering ETDs

Author

Feng Gu

Publication Date

9-5-2013

Abstract

Cloud computing paradigms are seeing very strong traction today and are being propelled by advances in multi-core processor, storage, and high-bandwidth networking technologies. Now as this growth unfolds, there is a growing need to distribute cloud services over multiple data-center sites in order to improve speed, responsiveness, as well as reliability. Overall, this trend is pushing the need for virtual network (VN) embedding support at the underlying network layer. Moreover, as more and more mission-critical end-user applications move to the cloud, associated VN survivability concerns are also becoming a key requirement in order to guarantee user service level agreements. Overall, several different types of survivable VN embedding schemes have been developed in recent years. Broadly, these schemes offer resiliency guarantees by pre-provisioning backup resources at service setup time. However, most of these solutions are only geared towards handling isolated single link or single node failures. As such, these designs are largely ineffective against larger regional stressors that can result in multiple system failures. In particular, many cloud service providers are very concerned about catastrophic disaster events such as earthquakes, floods, hurricanes, cascading power outages, and even malicious weapons of mass destruction attacks. Hence there is a pressing need to develop more robust cloud recovery schemes for disaster recovery that leverage underlying distributed networking capabilities. In light of the above, this dissertation proposes a range of solutions to address cloud networking services recovery under multi-failure stressors. First, a novel failure region-disjoint VN protection scheme is proposed to achieve improved efficiency for pre-provisioned protection. Next, enhanced VN mapping schemes are studied with probabilistic considerations to minimize risk for VN requests under stochastic failure scenarios. Finally, novel post-fault VN restoration schemes are also developed to provide viable last-gap recovery mechanisms using partial and full VN remapping strategies. The performance of these various solutions is evaluated using discrete event simulation and is also compared to existing strategies.

Keywords

network virtualization, virtual network mapping, virtual network survivability, cloud computing

Document Type

Dissertation

Language

English

Degree Name

Computer Engineering

Level of Degree

Doctoral

Department Name

Electrical and Computer Engineering

First Advisor

Ghani, Nasir

First Committee Member (Chair)

Ghani, Nasir

Second Committee Member

Shu, Wei

Third Committee Member

Pattichis, Marios

Fourth Committee Member

Shaban, Khaled

Share

COinS