My research deals with experimental software systems including operating systems, storage systems, and networked systems. My research focus and methodologies are rooted in the operating systems research community.
- Research summary (2008 PDF)
- Orca project
- Network/Internet Computing Lab (NICL)
- Selected recent publications
Updated March 2010
For the last decade or more I have been working toward a vision of an automated backbone of servers, storage, and other components managed as a shared platform for large-scale networked applications. Today, cloud is a popular term for networked hosting utilities, and commercial offerings are advancing rapidly. We are pursuing an ambitious cloud vision, viewing the server network and associated services as a critical public infrastructure requiring open, flexible, secure, robust, and decentralized control.
My research in this broad area focuses on providing operating system functions --- resource management, data storage and recovery, efficient and secure sharing, application packaging and launch, monitoring and adaptation, and so on --- at the granularity of a data center or multi-provider networked hosting system, rather than an individual computer. This work has spanned various subareas of systems research. For example, the services must sense-and-respond to adapt to changes in traffic demands or other conditions, while holding administrative burdens constant (self-managing or autonomic systems).
In the course of this research, we have constructed a family of experimental software prototypes for managing cloud hosting utilities with diverse substrate resources, multiple resource providers, and complex applications with varying requirements. One outcome of these efforts is the Open Resource Control Architecture (ORCA), an open-source platform deployed as a candidate control framework for NSF's Global Environment for Network Innovation (GENI). Much of my current work involves coordinated orchestration of dynamically switched optical networks and other virtualizable networks in conjunction with other cloud resources, under the GENI umbrella and based on the ORCA software, in collaboration with my colleagues at RENCI. I co-chair the Control Framework Working Group for GENI.
My published work has addressed the foundations of utility cloud computing: models to predict demand and behavior; virtualization and scheduling mechanisms to ``slice'' substrate resources as a measured and metered quantity; protocols to represent and enforce accountable contracts; control and optimization of resource slices for service quality metrics including dependability; scalable architectures for resource brokering and discovery; flexible and adaptive content services and network storage; data center architecture and server energy management; statistical inference of system behavior from instrumentation data; and various forms of dynamic resource management, including feedback-driven and market-based approaches.
A closely related focus is ``green'' computing, and in particular how ``smart'' electrical grids influence control policies in server clouds and other large energy consumers. My research portfolio includes the first papers on dynamic thermal management and adaptive energy management in hosting centers and other data centers. This body of work was also the first to consider how hosting centers could respond to congestion pricing for electrity, which is the cornerstone of demand-side management in emerging smart grids. Currently we are exploring the potential of energy-aware control policies in Duke's Scalable Computing Center, where I am faculty director. I am also interested in dynamic energy management for green buildings and power networks.
A second focus that emerged from our research in utility computing is how participants in distributed systems can be held accountable for faithful behavior. We introduced the notion of strongly accountable systems that enable each participant to determine if others are behaving correctly, without trusting assertions from another participant who may itself be faulty. For example, a key question for sustainable cloud economies is how providers can be held accountable for contractual commitments to their customers. More generally, accountability is a desirable property of social networked systems, e.g., for organizational cooperation in which users may attempt to deny their role in emergent outcomes.
Information about current research projects and lab facilities can be found through the Network/Internet Computing Lab (NICL).
I gratefully acknowledge support for this research from the National Science Foundation, IBM, and Network Appliance, and also (for earlier research) HP Labs and Cisco Systems.
Some old research can be found through the old Internet Systems and Storage Group (ISSG) site and the Network Storage Lab.