Human operators are an important aspect of any computing system or infrastructure; however, human errors in configuring systems pose reliability and security issues, which are increasingly serious as such systems grow ever more complex. In particular, numerous studies have shown that errors by human administrators have contributed significantly to misconfigurations of networks. The research community has reacted with development of solutions for improving traditional network infrastructure to identify and correct misconfigurations; however, such solutions reduce the impact of, rather than eliminate, misconfigurations. Automated workflows that abstract complex sets of administration tasks have also been seen as a potential solution. On the other hand, such tools simply remove the possibility of human error one step, to the development of the workflow, and can actually have the effect of magnifying the risk of such mistakes due to their speed of operation; also, the spectre of an administrator using the wrong tool continues to exist. Rather, the need is for a system that can examine the consequences of a proposed administrative action before it is taken.

In this project, I (under the advisory of Dr. Dutta) researched existing practices in network configuration and potential impacts of network misconfigurations. We proposed the design for SanityChecker – an SDN based proactive solution for intercepting
incoming configurations, inspecting them for human errors before configurations are committed to the devices, and flagging possible concerns. We developed a prototype integrated with OpenDaylight, a popular SDN orchestration product often adopted in industrial use, and invited apprentice and experienced real-world network administrators to work with it. Our results show that SanityChecker can successfully improve the operation of human network administrators.