FOR IMMEDIATE RELEASE CONTACT: December 21, 1993 Michael Schneider Pittsburgh Supercomputing Center 412-268-4960
The Heterogeneous Computing Challenge was held on Thursday, Nov. 18 at Supercomputing '93 in Portland, Oregon. Participating researchers and software developers used SCInet, the SC '93 exhibit-floor network, to demonstrate state-of-the-art heterogeneous processing. A team of judges representing university research, government laboratories and the computer industry reviewed 10 participating projects and issued awards in several categories, including most elegant and best speedup along with most heterogeneous.
Welling's project, "An Object-Oriented Distributed Ray Casting Volume Renderer," won most heterogeneous recognition because it used the greatest number of different-architecture computing platforms while at the same time being a "sensible" application. The application's user interface can spawn eight (with the potential for 64) separate rendering processes on a network of workstations. Each of these processes renders a portion of the dataset and returns a "subimage" to a compositing process that assembles the complete image and returns it to the user interface. The application uses object-oriented code (C++) and Parallel Virtual Memory (PVM) communication, which allows it to run on a wide range of platforms.
"In normal operations," said Welling, "the user interface would be on the user's workstation and worker processes would run on fast RISC machines, such as PSC's Alpha cluster." On the SC '93 showfloor, Welling ran the user interface on a Silicon Graphics Indigo workstation, using an SGI Indigo 2 for the compositing, with subimage processes running on the CRAY C90 back at Pittsburgh, a DEC Alpha and another SGI Indigo. "A relative lack of machines made it necessary to run multiple processes on each machine."
Ideally, said Welling, compositing would be done on a vector supercomputer like the C90, since this part of the operation is highly vectorizable. When PSC's 512 processor CRAY T3D system has full functionality, he added, this volume renderer could be distributed between the T3D, the C90 and a user workstation.
The impetus for the project, said Welling, grew out of his collaboration with Todd Elvins at San Diego Supercomputer Center. Welling also credited Pittsburgh Supercomputing Center staff members Wendy Huntoon and Jamshid Mahdavi for helping to get his application up and running on SCInet in Portland and Joe Lappa for helping to set up PVM.
The Pittsburgh Supercomputing Center, a joint project of Carnegie Mellon University and the University of Pittsburgh together with Westinghouse Electric Corporation, was established in 1986 by a grant from the National Science Foundation with support from the Commonwealth of Pennsylvania. Its purpose is to develop and make available state-of-the-art high-performance computing for scientific researchers nationwide.
go back to contents page