Abstract
|
The way datacenters and computer resources … The way datacenters and computer resources are managed has been changing, from bare metal servers and complex
deployment processes to on-demand cloud resources and applications.
The main technology behind this evolution was virtualization. By abstracting the hardware, virtualization decoupled software from the
hardware it runs on.
Virtual machine (VM) migration further increased the flexibility of management and maintenance procedures.
Tasks like maintenance, load balancing and fault handling were made easier.
Today, the migration of virtual machines is a fundamental tool in public and private clouds.
However as VMs rarely act alone, when the VMs migrate, the virtual networks should migrate too.
Solutions to this problem using traditional networks have several limitations: they are integrated with the devices and are hard to manage.
For these reasons the logical centralisation offered by Software-Defined Networking (SDN) architectures has been shown
recently as an enabler for transparent migration of networks.
In an SDN a controller remotely controls the network switches by installing flow rules that implement the policies
defined by the network operator.
Recent proposals are a good step forward but have problems. Namely, they are limited to a single data center or provider.
The user's dependency on a single cloud provider is a fundamental limitation. A large number of incidents involving
accidental and malicious faults in cloud infrastructures show that relying on a single provider can lead to the
creation of internet-scale single points of failures for cloud-based services.
Furthermore, giving clients the power to
choose how to use their cloud resources and the flexibility to easily change cloud providers is of great value,
enabling clients to lower costs, tolerate cloud-wide outages and enhance security. The objective of this dissertation is
therefore to design, implement and evaluate solutions for network migration in an environment of multiple clouds.
The main goal is to schedule the migration of a network in such a way that the migration process has the least
possible impact on the SDN controller's ability to manage the network. This is achieved by creating a migration plan that aims to minimize
the experienced control plane latency (i.e., the latency between the controller and the switches).
We have developed an optimal solution based on a linear program, and several heuristics.
Our results show that it is possible to achieve results close to the optimal solution, within reasonable
time frames. solution, within reasonable
time frames.
|