RIFFLE: A Distributed Network Operating System with Re-configurability for Multiple Domains

Boyang Zhou,Chunming Wu,Shuangxi Chen,Zhen Li,Hong Fan

DOI: https://doi.org/10.1109/iucc-cit-dsci-smartcns57392.2022.00030

2022-01-01

Abstract:The distributed network control in software-defined networking is greatly challenged by the design demands of evolving a running distributed control plane (DCP), and by the complexity of handling the control state dynamics. In this paper, we propose RIFFLE to address the problem using a distributed network operating system approach. RIFFLE enables the evolvability of control states of DCP deployed on a global scale. RIFFLE handles spatial and temporal dynamics that occur when updating and migrating the distributed control states, which overcomes the critical barrier to the evolvability. It fills the gap by enabling temporal reconfigurability in DCP. Moreover, RIFFLE reduces the complexities of building and maintaining network control services in DCP by enabling componentization and abstracting the underlying network dynamics. We evaluated the prototype RIFFLE through experiments running in PlanetLab. The results show that the prototype scales well for 135 nodes. We also validate that RIFFLE ensures the continuity and low content request delay for the supported the information-centric network supported during cache update periods owing to the enabled evolvability.