Matthew Caesar

Ph.D. University of California at Berkeley, 2007

Research Statement

The Internet is an extremely large and complex distributed system, composed of tens of thousands of competing ISPs and hundreds of millions of potentially-misbehaving hosts. Concerns about the Internet's ability to meet ever-increasing demands on performance and functionality in the presence of this complexity has led to a call to redesign the Internet's architecture, for example in the context of the NewArch project, and NSF's GENI and FIND programs. A key challenge faced in designing a new Internet lies in management and configuration. These issues were overlooked when designing early data networks and the Internet has been paying a massive price ever since. ISPs hire armies of engineers to manually configure routers and debug problems, and in daily life we are surrounded by an ever-increasing array of complex embedded devices that require substantial configuration to interoperate. Forcing humans to configure and manage networks increases reaction time to faults, introduces the potential for misconfiguration, and substantially increases operating costs.

What is lacking today is a principled look at how to make systems manage themselves. We need a fresh approach to designing networks and protocols with self-management in mind. Toward this goal my research focuses on protocols and systems that bootstrap, configure, and troubleshoot problems with only minimal manual intervention. In particular, these systems aim to self-configure in the presence of arbitrary topologies and failure modes, self-diagnose routing problems, and self-tune operation based on diagnoses.