Overlay and Peer-to-Peer Networks Research Projects in CMU

The OASIS group seeks to harness the collective computation, storage and communication resources of the millions of inexpensive and under-utilized personal computers distributed around the globe to build powerful distributed peer-to-peer systems. To realize this vision, we develop network technologies, architectures, protocols and algorithms that can enable the ubiquitous deployment of a wide class of emerging peer-to-peer applications. We place a heavy emphasis on realism, and our research activities have included collecting traces of actual applications running in peer-to-peer environments, and experiments on wide-area Internet test-beds.
End System Multicast
A wide range of group communication applications such as audio and video conferencing, multi-party Internet games, and distance learning are emerging today. Efficient network support for multi-point communication is a key requirement for the wide-spread deployment of these applications. The conventional wisdom has been that IP is the natural protocol layer for implementing multicast related functionality. However, ten years after its initial proposal, IP Multicast is still plagued with concerns pertaining to scalability, network management, deployment, and support for higher layer functionality such as error, flow and congestion control. In this project, we explore an alternative architecture where end systems implement all multicast related functionality including membership management and packet replication. We call such a scheme End System Multicast. This shifting of multicast support from routers to end systems has the potential to address most problems associated with IP Multicast. Yet, several issues need to be resolved before End System Multicast becomes a practical alternative to IP Multicast. We are studying these issues in the context of a protocol that we have developed: Narada. In Narada, end systems self-organize into an overlay using a fully distributed protocol, and optimize the efficiency of the overlay by considering application performance and adapting to network dynamics.

To demonstrate the feasibility of the End System Multicast architecture, we are broadcasting the ACM SIGCOMM Conference to be held in Pittsburgh between August 21-23, 2002 using End System Multicast.

Global Network Positioning (GNP)
Achieving high performance in peer-to-peer overlay applications in a scalable fashion is a challenging problem. Simple approaches that try to measure peer-to-peer network performance by brute force can be prohibitively expensive. In addition, many overlay algorithms require the peer-to-peer performance characteristics to be mapped onto a mathematical structure. Global Network Positioning (GNP) is a solution that can transform the complex peer-to-peer round-trip time relationships in the Internet into a simple geometric space structure (e.g. an N-dimensional Euclidean space). In this structure, each peer is characterized by a set of geometric coordinates that indicate its position. We have found through extensive Internet experiments that the geometric distances between peers in the GNP structure can accurately approximate the actual Internet network round-trip times. Thus the GNP structure can provide an accurate means for predicting Internet round-trip times, and provide a topologically sensitive mathematical structure for many overlay applications.

Peer-to-Peer Content Distribution
Content distribution on the Internet uses many different service architectures, ranging from centralized client-server to fully distributed. Compared to centralized schemes, fully distributed peer-to-peer content distribution provides more resilience and higher availability through wide-scale replication of content at large numbers of peers. However, wide-scale replication introduces many new challenges in locating and retrieving content. We design our content distribution system to deal with scale, heterogeneity and dynamic performance. We are involved in several ongoing projects that study different flavors of peer-to-peer content distribution. We explore the use of a simple, yet powerful observation called interest-based locality to provide scalable and high-performance content lookups and retrievals in peer-to-peer systems. We are also studying how to scale Gnutella, a popular file-sharing application. And, we are studying how selective use of peer-to-peer communications can enhance existing client-server systems in the CoopNet project.


Last modified: Wed Aug 14 16:12:11 EDT 2002