Abstract

An optical grid network reliably provides high speed communications. It consists of grid resources (e.g., computing and data servers) and huge-data paths that are connected to geographically dispersed resources and users.
One of the important issues is dimensioning optical grid networks, i.e., to determine the link bandwidth utilization and amount of server resources, and finding the location of servers. Another issue is the provisioning of the job requests (maximization of services) on the capacitated networks, also referred to as Grade of Service (GoS). Additionally, job scheduling on the servers has also an important impact on the utilization of computing and network resources. Dimensioning optical grid network is based on Anycast Routing and Wavelength Assignment (ACRWA) with the objective of minimizing (min-ACRWA) the resources.
The objective of GoS is maximizing the number of job requests (max-ACRWA) under the limited resources. Given that users of such optical grid networks in general do not care about the exact physical locations of the server resources, a degree of freedom arises in choosing for each of their requests the most appropriate server location.
We will exploit this anycast routing principle -- i.e., the source of the traffic is given, but the destination can be chosen rather freely. To provide resilience, traffic may be relocated to alternate destinations in case of network/server failures.

This thesis investigates dimensioning optical grids networks and task scheduling. In the first part, we present the link capacity dimensioning through scalable exact Integer Linear Programming (ILP) optimization models (min-ACRWA) with survivability. These models take step by step transition from the classical RWA (fixed destination) to anycast routing principle including shared path protection scheme. In the second part, we present scalable optimization models for maximizing the IT services (max-ACRWA) subject to survivability mechanism under limited link transport capacities. We also propose the link capacity formulations based on the distance from the servers and the traffic data set. In the third part, we jointly investigate the link dimensioning and the location of servers in an optical grid, where the anycast routing principle is applied for resiliency under different levels of protection schemes. We propose three different decomposition schemes for joint optimization of link dimensioning and finding the location of servers. In the last part of this research, we propose the exact task scheduling ILP formulations for optical grids (data centers). These formulations can also be used in advance reservation systems to allocate the grid resources.
The purpose of this study is to design efficient tools for planning and management of the optical grid networks.