Synthetic Forces Express

The Synthetic Forces Express (SF Express) project began in 1996 with the goal of investigating the use of high-performance computers as a means of supporting very large-scale Distributed Interactive Simulations (DIS). A specific charter for SF Express was to use supercomputers for demonstrating a scalable communications architecture, supporting tens of thousands of vehicles.

Over the past few years, SF Express has emerged as a distributed application harnessing multiple high-performance computational resources to meet the demands of large-scale network-based distributed simulations. SF Express has also become a "grid-enabled" application, able to incrementally incorporate services and translucent interfaces afforded by the evolving computational grid environment.

The first major milestone for SF Express was achieved in November 1996, with the successful simulation of more than 10,000 vehicles using the 1024-processor Intel Paragon. This demonstration gave us good insights for understanding inherant limitationss in the simulation engine, when dealing with dense scenario sets, as well as scaling issues we'd have to face dealing with distributed computing assets.

In 1997, SF Express was refined and extended to include multiple high-performance computers. A simulation of 50,000 vehicles was achieved using 1,904 total processors on six computers at sites distributed across seven time zones.

It soon became apparent that issues other than scalable communications needed to be addressed in order to improve the functionality and validity of large-scale distributed simulation experiments. Such issues include: scenario distribution, resource configuration, resource management, information logging, monitoring, and fault tolerance. To address some of these issues, we intergrated services provided by the Globus Metacomputing Toolkit.

On March 16, 1998, a record-breaking simulation was conducted using a 100,298 vehicle entity-level simulation - the largest, distributed, interactive battlefield simulation to date. This simulation was executed on 1,386 processors that were distributed over 13 computers among nine sites that spanned seven time zones. Global and local services were decoupled, allowing the application to run in a flexible, resource-aware environment. We continue to incorporate emerging computational grid tools and techniques into the distributed interactive simulation environment, bringing increased benefits of pervasive and dependable access to high-end computational capabilities.