Cornell University professors Keshav Pingali, Steve Vavasis, and Tony Ingraffea, and research associates Paul Stodghill, Gerd Heber, and Rob Cronin have successfully demonstrated a geographically-distributed simulation system, based on industry-standard Web Services, for solving coupled fluid/thermal/mechanical fracture problems.
"Grid computing is a metaphor representing many styles of distributed computing," said Cornell's India Professor of Computer Science Keshav Pingali, who is also an Associate Director at the Cornell Theory Center (CTC). "What we have shown is that some of the more useful styles of grid computing can be done quite effectively using existing industry-standard protocols and software such as SOAP and XML."
The path to discovery
The breakthrough was made in the course of implementing the Adaptive Software Project (ASP), a multi-institutional, multi-disciplinary computational science project, which is studying adaptivity in computational science applications. Researchers from the University of Alabama, Mississippi State University (MSU), Ohio State University, the College of William and Mary, and Clark-Atlanta University are partnering with Cornell in this project.
The benefits of the ASP approach of using industry-standard Web Services became evident while the team was exploring a simulation of a fracture in rocket engine components, such as those used in the space shuttle. These components transport high-pressure, high-velocity chemically reacting gases, which can create large thermo-mechanical stresses on component walls. To simulate fracture initiation and growth, the group had to integrate a number of large software systems, including a finite-element mesh generation code developed jointly by Cornell and the College of William and Mary, a chemically-reacting flow simulation code developed at MSU's Engineering Research Center and the University of Alabama, and a linear elastic fracture code developed at CTC.
"The traditional approach to integrating such software modules is to port all of them to a single computing platform," said Pingali. "Not only is this very time-consuming, but every time a new release of a module becomes available, some poor soul has to repeat the entire process of downloading and porting the code, re-compiling it, re-linking the compiled code with the rest of the software, and so on."
To simplify the job of integrating software components while respecting individual software developers' choices of hardware platforms, operating systems, and programming languages, the ASP team decided to deploy each major component as a Web Service running on a server at the institution where that component was developed. The flow simulation code for example runs on an IBM x330 Linux server at MSU, while the fracture simulation code runs on CTC's Windows cluster. The team uses industry-standard Web Service implementations such as Apache SOAP, and XML-based data exchange formats developed by Professor Steve Vavasis of the Cornell Computer Science Department.
"We view the person running the simulation as a client who writes a few hundred lines of code to invoke the various Web Services to orchestrate the simulation," said research associate Paul Stodghill, who wrote software for deploying legacy Unix codes as Web Services. "Our motto is 'write once, run from anywhere.'" He said that the overhead of using geographically-distributed Web Services for their simulation is about 10%.
Tony Ingraffea, the Dwight D. Baum Professor of Civil Engineering at Cornell and CTC Associate Director, feels that this overhead is worth paying. "Most applications people are not interested in using geographically-distributed computers to solve a large linear system, for example," said Ingraffea. "What we need is a way of building virtual organizations within which project members can work with each other's codes easily, while being sensitive to intellectual property issues."
Gordon Bell, senior researcher at Microsoft's Bay Area Research Center, concurs. "This project demonstrates the potential of a new way to build applications and the potential for a new software industry structure based on delivering results," he said. "Users don't have to buy apps programs and maintain a more complex software environment; they simply call a program or database. This is one of the few projects that I would call a Web Service, and it is well beyond what is running on today's experimental grid."
Frederica Darema, Senior Science and Technology Advisor in the CISE Directorate at NSF, is the cognizant NSF official for the ASP project. "It is to the credit of the scientists working on this project that they have developed such a cohesive collaboration," she said. "I am pushing for a new paradigm in application simulation and measurement capabilities called Dynamic Data Driven Application Systems, and the ASP model of multidisciplinary collaboration, together with the technology advances made by the project, are essential for enabling this new paradigm. I am very pleased with the outcomes of this project and its broader impact."
About the Cornell Theory Center
CTC is a high-performance computing and interdisciplinary research center located on the Ithaca campus of Cornell University with additional offices in Manhattan. CTC currently operates a Dell/Intel/Windows cluster complex consisting of more than 1500 processors, in addition to Unisys ES7000 Servers. Scientific and engineering projects supported by CTC represent a vast variety of disciplines, including bioinformatics, behavioral and social sciences, computer science, engineering, finance, geosciences, mathematics, physical sciences, and business. For more information, visit http://www.tc.cornell.edu or http://www.ctc-hpc.com.