Graduate Students:
G. Rob Malan,
Robert Hall,
Craig Labovitz,
and Jimmy Wan
Research Scientists: Sushi Subramanian and Peter Knoop
Professors: Farnam Jahanian and Atul Prakash
Sponsor: National Science Foundation, Intel Corporation
This project is currently sponsored by the National Science Foundation and a gift from the Intel Corporation. The primary objectives of the proposed research are: (1) the development a suite of scalable middleware services which can adapt to network and client variability to support wide-area applications such as webcasting and group collaboration; and (2) the construction of a distributed testbed to demonstrate this middleware and to evaluate various design tradeoffs through experimental studies.
The availability of ubiquitous network connections in conjunction with significant advances in hardware and software technologies have led to the emergence of a new class of distributed applications. This research focuses on the systems support needed for two of these applications: group collaboratories and webcasting. The problem of data transport in a geographically dispersed wide-area collaborative environment has two inherent properties: (1) the data is often shared under real-time conditions, and (2) the computing resources available to the collaboratory's participants vary considerably. The first is an artifact of the collaboration process. The second reflects the heterogeneity of network connectivity in wide-area networks such as the Internet: there will always be slower members that cannot keep pace with faster, better-connected participants.
Our data dissemination service, called Salamader , is based on a push architecture. The approach we advocate is the use of Semantic-based Application-level Quality of Service policies. These policies are a high-level approach to mapping semantic knowledge about a collaboration activity, the degree of consistency of data shared and the roles of participants, into quality of service policies for network and computational resources. Instead of reserving or acquiring a guaranteed rate of service from the network, it acknowledges the fact that it may be impossible to achieve a homogeneous guarantee for all of the participants. We have introduced a software architecture for supporting these policies by means of plug-in modules that can be dynamically placed at all points in the collaboratory's datapath. These modules fall into one of the following three categories: delivery, transformation, and synchronization. When used in concert, these modules provide a flexible framework for implementing application-specific semantic-based quality of service policies.
The prototype software from this research is currently being used to support collaborative computing for space science campaigns in the Upper Atmospheric Research Collaboratory (UARC) project, a multidisciplinary research effort to develop an experimental testbed for examining issues is supporting collaborative scientific work over wide-area networks. This software is also being used as part of Merit Network's Internet Performance Measurement and Analysis (IPMA) project for webcasting of the Internet loss and latency measurements at the major public exchange points.
We are currently constructing a testbed to demonstrate the scalability of our data dissemination software and to evaluate various design tradeoffs through experimentation. An Intel-based testbed at the University of Michigan will be connected to other wide-area collaboratories to perform these studies. This project will be able to exploit two kinds of network connections to sites outside our university: an experimental high-speed low-latency link (vBNS) and the traditional connection to the Internet infrastructure. This mixture will provide us with a unique testing ground for the scalability of future wide-area collaboratories.
farnam@eecs.umich.edu