Automated profiling framework for software-defined networking

Abstract

Software-Defined Networking (SDN), which separates the control and data plane of network, is strongly considered as a promising future networking architecture. Compared with legacy networking archi-tecture, it allows to enable a variety of innovative network functions at much less cost and effort. According-ly, each component of SDN is also being rapidly realized, and one of the most noticeable SDN component implementations would be SDN controllers, such as ONOS or Floodlight. One advantage of these SDN con-trollers is capability of hosting various network applications to enable innovative network functions

however, it is crucial to analyze these applications before the actual deployment as they may directly affect the per-formance of the managed network. To be more specific, SDN applications may contain performance bugs that unnecessarily consume significant system resource or produce critical bottlenecks in the controller. In this paper, we introduce an automatic SDN application profiling framework, SPIRIT, which reduces the human effort in revealing any performance bugs that might exist in SDN applications. In order to show the effective-ness of our framework, we reveal new performance bugs exist in ONOS and Floodlight applications.