With new solutions for application development, developers can work with familiar tools in a single operating environment, reducing the complexity and cost normally associated with HPC.
Most developers find it complex, time consuming and expensive to develop separate applications for desktop or single server computing and high-performance scale-out cluster or grid environments. As a result, companies are demanding a single, seamlessly integrated, easy to use environment for all their computing needs, whether it is word processing or seismic analysis.
“In the past, developing a scientific or engineering application solely for a tightly-coupled cluster was acceptable, because dedicated, isolated clusters were the most viable option for high-performance computing,” says John Powers, President of Digipede Technologies, an HP partner. “But now, there are scores of mainstream application workloads in enterprise datacenters and on desktops that can benefit from high-performance computing resources. We need to give mainstream developers easy, familiar tools so they can take advantage of those HPC resources.”
Developer needs
Martin Schultz, Founder of Scientific Computing Associates and Arthur K. Watson, Professor of Computer Science at Yale University, agree that many developers want to take advantage of new, high-performance cluster technologies. And they want to build on the familiar Windows® platform. “Since the vast majority of mainstream applications is not parallel (high-performance) and are built for Windows, developing a separate set of high-performance applications creates a host of complications,” Schultz explains. “More and more, developers want to work with Windows® Compute Cluster Server (CCS) 2003 because of the familiar development environment and inherent linkage with standard applications and desktop systems.”
Windows CCS offers a seamless operating environment that spans both front-end and back-end systems, eliminating the need to develop separate “computing” and “high-performance computing” applications. What’s more, development tool providers have come out in force to help developers build or port applications to Windows CCS. For example, high-performance parallel applications formerly required message passing interface (MPI) protocols. With more than 300 commands, MPI is complex, and there are few experts versed in its usage. Development tools such as Scientific Computing Associates’ Linda (precompiler), however, offer a simplified means—with a crisp six commands—to develop high-performance parallel applications quickly and cost-effectively.
“Most developers have avoided high-performance applications because of the time, complexity and integration hurdles involved,” claims Schultz. “Microsoft has a long history of offering high-level, easy to use tools for users. We have taken this philosophy to heart, creating a simplified alternative that allows developers to create applications easily for use among multiple, multi-core nodes. High performance is obtained in days, not months or years.”
Another HP partner, The Portland Group, offers parallel Fortran, C and C++ compilers for 64-bit x64 and 32-bit x86 processor-based workstations, servers and clusters. The compilers work hand-in-hand with development tools such as Linda, says Doug Miles, Director of the Portland Group. “Whether a developer is creating a new application from scratch or porting an existing application from Linux® or UNIX® to Windows CCS,” he says, “the process of high-performance, grid-based application development is much, much easier than it has been in the past.”
Windows CCS development tools
Digipede Technologies is also an enthusiastic participant in the Windows CCS development frenzy, having created a solution that helps distribute Microsoft® .NET application workloads across many machines. “The Digipede Network is the first commercially available distributed computing solution to take advantage of the speed, security and interoperability of .NET,” notes Powers. “It allows developers to continue making use of the tools they know best, while enabling them to extend their applications within a grid environment for greater scalability and speed.”
And for those wanting to develop highly advanced parallel applications that employ sophisticated message passing techniques or migrate from other existing environments, HP offers HP-MPI, a robust and streamlined version of MPI. “There is a full range of development tools for those seeking to build high-performance applications,” says Betty Van Houten, HP-MPI Research and Development Project Manager for HP. “Whether a company wants to save development costs and quickly create a parallel application using Linda or construct a highly refined, long-term solution using HP-MPI, there are a multitude of good options.”
The division between everyday computing and high-performance computing is becoming a myth, says Cox. “Developing for clustered or distributed grid systems is no longer complex, time-consuming or splintered among different operating systems,” he suggests. “Developers can use the tools they love, work within a single operating environment, reduce their development costs and pass along the benefits to their customers.”
|
Printable version
