A SDN-based network architecture for cloud resiliency

In spite of their commercial success, Cloud services are still subject to two major weak points: data security and infrastructure resiliency. In this paper, we propose an original Cloud network architecture aiming at improving the resiliency of Cloud network infrastructures interconnecting remote datacenters. The main originality of this architecture consists in exploiting the principles of Software Defined Networking (SDN) in order to adapt the rerouting strategies in case of network failure according to a set of requirements. In existing Cloud networks configurations, network recovery after a fiber cut is achieved by means of the usage of redundant bandwidth capacity preplanned through backup links. Such an approach has two drawbacks. First, it induces at a large scale a non-negligible additional cost for the Cloud Service Providers (CSP). Second, the pre-computation of the rerouting strategy may not be suited to the specific quality of service requirements of the various data flows that were transiting on the failing link. To prevent these two drawbacks, we propose that CSPs deploy their services in several redundant datacenters and make sure that those datacenters are properly interconnected via the Internet. For that purpose, we propose that a CSP may use the services of multiple (typically two) Internet Service Providers to interconnect its datacenters via the Internet. In practice, we propose that a set of "routing inflection points" may form an overlay network exploiting a specific routing strategy. We propose that this overlay is coordinated by a Software Defined Networking-based centralized controller. Thus, such a CSP may choose the network path between two datacenters the most suited to the underlying traffic QoS requirement. The proposed approach enables this CSP a certain independency from its network providers. In this paper, we present this new Cloud architecture. We outline how our approach mixes concepts taken from both SDN and Segment Routing. Unlike the protection techniques used by existing CSPs, we explain how this approach can be used to implement fast rerouting strategy for inter-datacenter data exchanges.