Schooner Project Research Exhibit Proposal

To: Wilf Pinfold, SC'95 Research Exhibits Chair

From: Patrick T. Homer

Date: July 31, 1995

The Schooner Project, a part of the Department of Computer Science at The University of Arizona, proposes a research exhibit entitled The Schooner Interconnection System at Supercomputing '95. This proposal includes a brief abstract that can be used to publicize our exhibit, all exhibit logistics, and a description of the technical projects we intend to exhibit at SC'95.

Contact person for this exhibit:

Exhibitor names (tentative):

Abstract

The Schooner Project is investigating the use of software interconnection systems in the design of scientific meta-computations. Meta-computations are built from heterogeneous collections of applications that cooperate in the solution of a problem. An interconnection system must solve the dual problems of establishing communications among the applications and providing configuration tools that allow flexible and dynamic mappings of components onto hosts. The Schooner system provides such tools in the context of scientific applications and is being used with several meta-computations that will be demonstrated at Supercomputing '95. The meta-computations are on-going projects developed through collaborations with the NASA Numerical Propulsion System Simulator (NPSS) project and an NSF-funded Ecosystem Modeling project. The demonstrations include a jet engine simulation that combines one- and three-dimensional engine component simulations; and a high performance ecosystem meta-computation combining visualization, parallel simulation, and remote GIS database applications. These meta-computations include components developed in a variety of scientific programming languages, and use several different programming models. They will employ a variety of machines on the exhibit floor and remote machines distributed over a wide geographic area.

Exhibit Logistics

Equipment:
At this time we anticipate that our exhibit will be using four workstations: 

	1  Sparc 10
	2  Dec Alphas
	1  SGI
Size and Layout:
We request an exhibit area to include 1 x 2 blocks -- a 10' x 20' booth.
A preliminary layout of how we plan to organize this area:

Exhibit SciNet Connections:
This information is essential! Give specific details of connections needed from SciNet. The default connection to each exhibit is a simple 10baseT ethernet drop.

We will need IP numbers for each of the 4 workstations. We will bring a multi-port repeater. It will connect to the default 10baseT ethernet drop into the booth and to the 4 workstations.

Exhibit Furnishings:
A set of two chairs and two 2x6 or 2x8 tables is provided for each block (10x10 feet) in your exhibit area. Please comment about any additional furniture you will need and how you propose to acquire it.

The default tables and chairs will be adequate. We will have a few poster displays to set up behind the tables and will bring supports for them.

Exhibit power connections:
A single 110-volt power box is provided for each block (10x10 feet) in your exhibit area. If you anticipate unusual power requirements, provide an estimate of what your needs will be.

The default power supply will be adequate.

Shipping:
Is any special help requested?

We intend to bring a van with our equipment in it. We only need assistance in getting the equipment from the van to the exhibit booth.

Exhibit Technical Plan

The research exhibit will demonstrate Schooner's ability to facilitate the construction of scientific meta-computations. The exhibit will focus on our collaborative research in two on-going projects. The demonstrations will include the workstations within the booth, and a variety of remote platforms located at NASA Lewis Research Center, the University of Toledo, and The University of Arizona. In addition to the active demonstrations, we will have displays and handouts that describe Schooner and the projects in which it is currently involved.

Three of the demonstrations are part of on-going, interdisciplinary collaborations:

Prototype simulation executive. The executive uses the Network Editor of AVS to create a one-dimensional jet engine model from a palette of engine component modules. A number of the engine modules have been extended using Schooner to allow the computation portion to execute on a variety of remote machines. The demonstration will involve a number of the modules executing their computational portions on a suite of remote machines, and highlight the ability to dynamically start, stop, and move the remote computations among the available computational platforms.

Prototype zooming meta-computation. The one-dimensional jet engine model has been extended to incorporate ADPAC, a fully three-dimensional flow analysis package, to simulate the fan component of the engine. Up to eight instances of ADPAC can be executed remotely, and the results used to create a single-curve performance map that is interpolated by the one-dimensional code. A monitoring tool, constructed with TAE+ and executing on a second workstation, is used to watch the progress of the fan code instances. It provides continuous monitoring of the convergence characteristics of each ADPAC instance. An expert system is used to report various warnings to the user.

Ecosystem modeling. The demonstration will include a graphics tool that allows the user to steer and to see the results from a discrete event simulation running on a remote parallel platform. The remote simulation executes on either a cluster of workstations or a CM-5. Both the simulation and the visualization tool access a remote Geographic Information System (GIS) database executing on a Sun Sparc shared-memory multi-processor.

The first two demonstrations are part of the NPSS project at NASA Lewis Research Center, and are collaborations with researchers at the University of Toledo and Cleveland State University. The third demonstration is part of the High Performance Simulation group that includes researchers at The University of Arizona in the Electrical and Computer Engineering Department and the School of Renewable Natural Resources.

Finally, we have several smaller demonstrations that illustrate Schooner's ability to connect to a variety of remote computations and that illustrate its flexibility and dynamic configuration features. One example in particular uses a neural net code that executes on a variety of remote parallel platforms:

Schooner provides the connections between the remote computation and the AVS visualization system, and allows the user to dynamically select the remote platform and steer the neural net during its execution.
Queries to Patrick Homer, patrick@cs.arizona.edu