Consistent updates in software defined networks: On dependencies, loop freedom, and blackholes

We consider the problem of finding efficient methods to update forwarding rules in Software Defined Networks (SDNs). While the original and updated set of rules might both be consistent, disseminating the rule updates is an inherently asynchronous process, resulting in potentially inconsistent states. We highlight the inherent trade-off between the strength of the consistency property and the dependencies it imposes among rule updates at different switches; these dependencies fundamentally limit how quickly the SDN can be updated. Additionally, we consider the impact of resource constraints and show that fast blackhole free migration of rules with memory limits is NP-hard for the controller. For the basic consistency property of loop freedom, we prove that maximizing the number of loop free update rules is NP-hard for interval-based routing and longest-prefix matching. We also consider the basic case of just one destination in the network and show that the greedy approach can be nearly arbitrarily bad. However, minimizing the number of not updated rules can be approximated well for destination-based routing. For applying all updates, we develop an update algorithm that has a provably minimal dependency structure. We also sketch a general architecture for consistent updates that separates the twin concerns of consistency and efficiency, and lastly, evaluate our algorithm on ISP topologies.