Achieving Memory-saving Network Update

Jiyang Liu, Liang Zhu, Weiq

Abstract

Software defined networking (SDN) offers opportunities to develop high-level abstractions for implementing network update, but current SDN controller platform lacks effect mechanisms for updating network configuration on the fly. There exist two main challenges for implementing network update: 1) network is a distributed system and 2) network controller can only update one network node at a time. Naïvely updating individual nodes may lead to incorrect network behaviors. Elegant solution based on two-phase update can guarantee that traffic will be processed consistently during network update, which means each packet can be routed based either on initial network configuration or target network configuration, but never a mixture of the two. Implementing consistent network update is expensive based on previous approach, and we present mechanisms for memory-saving two-phase network update. Our design addresses one major problem: how to delete the initial configuration effectively from network nodes when controller enforces the target configuration. We propose a hierarchical network metadata structure to accelerate the procedure of removing the initial configuration. Finally, we describe the results of some experiments demonstrating the effectiveness of configuration deletion and the effect of memory-saving for the network.

Relevant Publications in Information Technology & Software Engineering