Transparent data plane failure tolerance for software defined networks

With the ever-increasing size and scale of networks and the critical nature of user applications, it is very important to have adequate failure tolerance mechanisms in order to ensure minimal disruption is user operations. Rather than handling failures in the control plane, it is more desirable to have failure tolerance mechanisms in the data plane. To this end, Software Defined Networking (SDN) has introduced different measures for data plane failure tolerance in existing protocols. More specifically in OpenFlow, this is achieved through fast fail-over groups that monitor health of ports and route traffic along alternative ones when the primary port is unavailable. However, configuring fast fail-over groups in data plane has required all logic to be off loaded into controller applications. This not only makes these applications more complicated, but has also proven to be restrictive in terms of the application's architecture and programming constructs. In this thesis we introduce a novel design for SDN that abstracts out the failure tolerance logic from the control plane and automatically and transparently configures fast fail-over groups in the data plane, while leaving controller applications unmodified. Our technique pre-computes backup paths for each link in the data plane and installs them as MPLS tunnels. It then configures and installs fast fail-over groups to use these tunnels for rerouting traffic in the event of a link failure. Finally, it transparently intercepts communication between the control and data planes to use these fast fail-over groups and translate any other messages back and forth as needed. We also provide a module to make failure tolerance compatible with controller applications running in different versions of OpenFlow, particularly OpenFlow version 1.0, which has no notion of fast fail-over groups and integrates seamlessly with our design to make non-resilient controller applications failure tolerant. Our experiments demonstrate that our design is effective in transparently configuring fast fail-over groups for non-resilient controller applications, achieves a reasonably high throughput and incurs a fairly low performance overhead on network resources.LimitedAuthor requested closed access (OA after 2yrs) in Vireo ETD syste