RESEARCHERS:
Fred Annexstein fred.annexstein@uc.edu
Kenneth
Berman
ken.berman@uc.edu
Raj Bhatnagar raj.bhatnagar@uc.edu
Chia-Yuan
Han chia.han@uc.edu
John Schlipf
john.schlipf@uc.edu
LINC is a testbed high-speed research network of computer workstations that is being developed using funding from the National Science Foundation MRI program and matching funding from the University of Cincinnati. This testbed supports the needs of a variety of distributed computing system and network protocol research projects. The testbed currently includes a cluster of high-performance workstations locally connected by a high-speed LAN, a Beowulf cluster of 64 Pentium III processors and a second Beowulf cluster of 32 Athlon MP processors, both running linux, with connectivity soon to be provided to the statewide Cluster Ohio Project. The projects supported by LINC exploit fundamental interconnections between theory, systems and applications of distributed computing, and include:
Validation of Communication and Routing Schemes in Distributed Computing Environments: We focus on the algorithmic support required to efficiently and reliably implement communications and communication services in distributed networks. We focus our efforts on (1) supporting communication through adaptive multi-tree routing schemes, (2) designing efficient label-correcting algorithms, (3) developing heuristic algorithms for dynamically building and maintaining constrained multicast trees, and (4) testing, through simulation on the experimental test-bed, the performance of the designs and algorithms developed in this project.
Optimistic Compiler Optimizations: We study compiler optimization techniques based on program analysis and partitioning and scheduling on a target architecture consisting of heterogeneous cluster of workstations connected by a heterogeneous network. We propose to devise a new set of optimizations called as optimistic profile guided optimizations that would result in more efficient program partitioning.
Distributed Decision Support and Data Mining: We study how to adapt decision support, data mining, and machine learning algorithms, designed for data residing on a single computer system, when the data resides on distributed computers and cannot be transported freely. A major goal of our continuing research is to determine which inference and learning mechanisms can be decomposed for implementation in such distributed knowledge environments. Automated problem dependent decomposition and increase in complexity of these algorithms due to excessive message exchanges are the focus of our current research.
Development of a Collaborative Computing Environment: We study system implementation issues arising from the support of multimedia-based collaboration in distributed environment. The research will focus on two major thrusts, i) system architecture, and ii) human-computer interface.
Heuristic Search Strategies on Beowulf Clusters: We study a certain hybrid version of standard parallel backtracking used in combination with various heuristic search strategies. Our search strategies are implemented using the MPI parallel programming language on a Beowulf cluser, but the strategies can also be adapted to other parallel languages. Good speedup is obtained for problems modeled on state space trees by using repartitioning techniques to minimize the number of idle processors.
In addition to supporting the above NSF-sponsored research projects, LINC is affliated with and supports research in the following laboratories: