Distributed control plane for software-defined networks
Sukanya Bhowmik, Muhammad Adnan Tariq, Boris Koldehofe, André Kutzleb, Kurt Rothermel · 2015
Realizing a communication middleware in a software-defined network can leverage significant performance gains in terms of latency, throughput and bandwidth efficiency. For example, filtering operations in an event-based middleware can be performed highly efficiently in the TCAM memory of switches enabling line-rate forwarding of events. A key challenge in a software-defined network, however, is to ensure high responsiveness of the control plane to dynamically changing communication interactions. In this paper, we propose a methodology for both vertical and horizontal scaling of the distributed control plane that is capable of improving the responsiveness by enabling concurrent network updates in the presence of high dynamics while ensuring consistent changes to the data plane of a communication middleware. In contrast to existing scaling approaches that aim for a general-purpose distributed control plane, our approach uses knowledge of the application semantics that is already available in the design of the data plane of a communication middleware, e.g. subscriptions and advertisements in an event-based middleware. By proposing a methodology for an application-aware control distribution, we show, in the context of PLEROMA, an event-based middleware, that application-awareness is significantly beneficial in avoiding the synchronization bottlenecks for ensuring consistency in the presence of concurrent network updates and thus greatly improves the responsiveness of the control plane.