Network Verification and Fault Diagnosis in SDN
Associate Professor in School of Information Technology Illinois State University. His research interests are in fault diagnosis in large-scale distributed networking systems, cloud computing, network security, wireless networks, and traffic classification, as well as various machine learning and AI techniques.
ABSTRACT：Compared to traditional networks, SDN employs two important simplifications that make it eligible for another new dimension of network troubleshooting: formal verification. First, SDN relocates control from distributed algorithms running on individual devices to a logically centralized controller. Second, SDN replaces various heterogeneous devices (e.g., switches, routers, load balancers, firewalls) used in traditional networks with commodity programmable SDN switches that provide a standard set of features. Together, these simplifications imply that the behavior of the network is determined by the sequence of configuration instructions issued by the controller. Hence, network operators can reason about the states of SDN switches to verify whether or not the network has some property (e.g., loop free, correct access control). In the networking community, there is burgeoning interest in tools that check network-wide properties automatically. In this talk, I will start from a traditional fault diagnosis technique based on Bayesian belief network, and then move to two dimensions of network troubleshooting in the context of SDN: causality based fault diagnosis and formal method based network verifications. Several recently developed solutions in the networking community, such as Minimal Causal Sequence (MCS), FlowChecker, Header Space Analysis (HSA), VeriFlow, will be presented as case studies. Finally, I will wrap up my talk with brief discussion of my recent research in the related field.