Systems Seminar: Peter Druschel, Rice University

Structured Peer-To-Peer Overlay Networks: A New Foundation For Distributed Applications?

Abstract

Peer-to-peer (p2p), initially conceived for the purpose of sharing music in the Internet, is emerging as a much more general paradigm for the construction of resilient, large-scale, distributed services and applications. We define p2p systems broadly as self-organizing, decentralized distributed systems tha consist of potentially untrusted, unreliable nodes with symmetric roles. The scalability and resilience of p2p systems lends itself to a growing domain of applications beyond file sharing. At the same time, their scale, decentralization, diversity and lack of trust also pose difficult problems, particularly in resource management and security.

Recent work on structured p2p overlay networks like CAN, Chord, Pastry and Tapestry has made significant strides towards providing a general substrate that simplifies the construction of robust, large-scale distributed applications. These overlays effectively shield application designers from the complexities of organizing and maintaining a secure overlay network, tolerating node failures, balancing load, and locating application objects.

In this talk, I'll present an overview of the state-of-the-art in structured p2p overlays that provide self-organization, fault-tolerance, efficient object location, and proximity-aware overlay construction. I will also sketch the design of several applications, including cooperative network storage, scalable endsystem multicast, and content distribution. I'll conclude with an outlook on key research problems and future directions.

Speaker Bio

Peter Druschel's research interests are in operating systems, networks, and computer architecture. Currently, his research focuses on providing operating system support for high-speed networking. Fundamentally, the role of an operating system is to abstract the physical resources of the underlying hardware and to multiplex these resources among multiple application programs and users. In this role as a mediator between application programs and underlying hardware, modern operating systems are being challenged on both fronts. On the one hand, operating systems are expected to efficiently support a broad range of application programs, including scientific, engineering, business, and entertainment applications. On the other hand, a single operating system is expected to efficiently run on a wide range of computer systems, from programmable data assistants and personal computers to parallel supercomputers, all connected by networks with widely differing performance characteristics. Druschel's long-term research goal is to develop an operating system that can be configured to run on radically different distributed hardware platforms, the services of which can be dynamically customized and extended to efficiently support specific applications. Toward this end, his research focuses on communications-oriented operating system design and application-specific customization of operating system services. In 1995 Druschel received an NSF career grant for research in "Communication-Oriented Operating Systems for the National Information Infrastructure."