CCU: Algorithm for Concurrent Consistent Updates for a Software Defined Network

Abstract

A Software Defined Network has its control plane separated from its data plane, with the control plane providing abstractions for development of control programs that alter the state of the network by updating the rules installed in switches. Switch updates are central to an SDN and updates must be such that there are no packet drops or loops. An update property which ensures that the traces generated for a packet are either due to the old network configuration or due to the new configuration, but never a combination of both, thus preventing packet drops and loops, is called per-packet consistency. We envisage that large data center networks supporting multiple tenants and network virtualization will need a large number of concurrent updates due to VM creation, VM migration, network management applications etc. There are algorithms in the literature that enable concurrent updates and improve update speed but they are either not general or do not preserve per-packet consistency. This paper proposes an update algorithm, Concurrent Consistent Updates (CCU), that is general, is per-packet consistent and that enables concurrent disjoint updates. Since the size of a TCAM on a switch is small, rule tables implemented in software are used to supplement the TCAM. The algorithm makes use of a Software Rules Table to restrict updates to only the internal switches that genuinely need an update and to reduce update installation time.