Provable data plane connectivity with local fast failover: introducing openflow graph algorithms

Abstract

Modern software-defined networks support the implementation of in-network failover mechanisms: mechanisms to quickly re-establish connectivity in the data plane without the interaction of the software controller. Interestingly, however, not much is known today about how to make use of these mechanisms. This paper shows a very strong result: there exist failover implementations for OpenFlow that achieve a maximal robustness, in the sense that connectivity is always ensured as long as the underlying physical network is connected. In particular, we show that the problem of computing failover tables is related to graph search, and present three different algorithms achieving different tradeoffs, in terms of the number of required failover rules, the number of tags, as well as the resulting path lengths. Our work can also be seen as a first attempt to implement classic graph algorithms in OpenFlow.