Abstract

Optical networks play a crucial role in the development of Internet by providing a high speed infrastructure to cope with the rapid expansion of high bandwidth demand applications such as video, HDTV, teleconferencing, cloud computing, and so on. Network virtualization has been proposed as a key enabler for the next generation networks and the future Internet because it allows diversification the underlying architecture of Internet and lets multiple heterogeneous network architectures coexist.

Physical network failures often come from natural disasters or human errors, and thus cannot be fully avoided. Today, with the increase of network traffic and the popularity of virtualization and cloud computing, due to the sharing nature of network virtualization, one single failure in the underlying physical network can affect thousands of customers and cost millions of dollars in revenue. Providing resilience for virtual network topology over optical network infrastructure thus becomes of prime importance.

This thesis focuses on resilient virtual topologies in optical networks and cloud computing. We aim at finding more scalable models to solve the problem of designing survivable logical topologies for more realistic and meaningful network instances while meeting the requirements on bandwidth, security, as well as other quality of service such as recovery time.

To address the scalability issue, we present a model based on a column generation decomposition. We apply the cutset theorem with a decomposition framework and lazy constraints. We are able to solve for much larger network instances than the ones in literature. We extend the model to address the survivability problem in the context of optical networks where the characteristics of optical networks such as lightpaths and wavelength continuity and traffic grooming are taken into account.

We analyze and compare the bandwidth requirement between the two main approaches in providing resiliency for logical topologies. In the first approach, called optical protection, the resilient mechanism is provided by the optical layer. In the second one, called logical restoration, the resilient mechanism is done at the virtual layer. Next, we extend the survivability problem into the context of cloud computing where the major complexity arises from the anycast principle. We are able to solve the problem for much larger network instances than in the previous studies. Moreover, our model is more comprehensive that takes into account other QoS criteria, such that recovery time and delay requirement.