The Cornell Theory Center (CTC) is an HPC and interdisciplinary research centre located on the Ithaca campus of Cornell University with additional offices in Manhattan. Before 1997, as Gerd Heber stated, CTC was one of the supercomputing centres of the United States National Science Foundation (NSF). The systems in use were the KSR1 with 128 processors purchased in 1992, the SP1 with 64 processors installed in April 1993 and the SP2 with 512 processors implemented in October 1994.

In 1997, CTC received a grant from Intel for the first cluster based on Windows NT 4.0 and provided with MPI software technology MPIPro and Cluster CoNTroller. In 1999, as the speaker told the audience, on August 3, a V1 cluster followed with 64 Dell PowerEdge servers, 4 x Pentium III@500 MHz, 4GB RAM per node, 2MB L2 cache per processor, Giganet cLAN interconnect, Windows NT 4.0 Enterprise Edition, and 47 GFlops. This system ranked number 381 in the Top500 in June 2000.

Gerd Heber continued to explain how the Crack Propagation on Teraflop Computers (CPTC) programme, funded by NSF-CISE, was ported to Windows NT 4.0 server. This programme consists of 3 mesh generators; FemLib; preconditioners; H/P adaptivity, third party packages such as PETSc, BlockSolve95, ParMetis, DSCPACK, PSPASES, and JANUS. It has about 1.5 million lines of code. The main goal was to preserve the built-infrastructure. The porting included three months, 80 percent of infrastructure, and 20 percent of code.

Since 2000, the following tools have been developed: the Windows 2000 Server 64-bit edition, the Microsoft SQL "Yukon" Server 2000 64-bit edition, the Intel Itanium processor, the Intel C++/Fortran OpenMP compilers, IntelVTune Performance Libraries, .NET Framework (1.0, 1.1., 2.0), Visual Studio .NET 2003, and Windows Server 2003.

The current infrastructure at CTC includes clusters, a tape robot, database servers, an active directory, a scheduler, a certificate server, login nodes, a cave, file servers, an SCTS, and Web servers. Gerd Heber noted that some important pieces are still obviously missing such as the parallel debugger, a decent shell, a parallel file systems, and an MPI library and start-up out-of-the box but the speaker is expecting some major announcements from Microsoft.

Windows stands for an integrated HPC solution and the reason why is that two technologies have changed the way HPC is done nowadays. These technologies are databases and Web services. The first one is provided for by Windows with the SQL server 2000/5 and the second one with the .NET Framework.

In visual SQL the goal is to develop a real time interactive visualization and steering environment for multi-scale and multi-physics simulations, as Gerd Heber told the audience. The application is based on OpenDX and SQL Server. The collaborators at Microsoft Research are Jim Gray and Todd Needham.

At CTC, an adaptive software project for field-driven simulations is now running funded by NSF-ITR. The aim is to implement a system for multi-physics, multi-scale adaptive CSE simulations using computational fracture mechanics and chemically-reacting flow simulation. The project will allow the researchers to understand the principles of implementing adaptive software systems, as the speaker explained.

The adaptivity in CSE simulations is playing at three levels. At the level of the applications, there is a change in modelling and governing equations. Gerd Heber mentioned the elasticity of PDEs versus molecular-scale interactions as an example. At the level of the algorithms, there is a change in the solution method for governing equations, illustrated with the difference between finite element and wavelet bases. At the level of the system, there is a response to the changing of the resource availability. As an example, the speaker cited the processor/link failure.