High Performance Computing and Nuclear Engineering

Until the middle of the XXth century, scientific and technological researches were based on two methodologies: theoretical and experimental. With the arrival of the computer a third one emerged: computing methodology, which opened its own perspectives of development, complementing and supplementing the traditional research methodologies.

It is not surprising that the development of the computer, which since its appearance has been showing impressive speed and memory improvements, resulted in great impact in the scientific and technological development processes. In the computing mechanics field, for example, the onset of supercomputing, achieved through parallel processing, allows that almost complete projects be developed in the computer: virtual prototypes before the real prototypes. It is unnecessary to emphasize the importance of that to meet the technological innovation speed demanded by society and market.

While computing methodology is being developed vertiginously, it is worth to remember that power reactors appeared in the sixties, before the maturing of computers and computing methods as powerful tools for analysis and design. PWR technology developed from simplified models, needing great investments in experimental facilities. The results of the huge experimental effort involved in the development of PWR reactors constitute property and industrial secret of companies like Westinghouse and KWU, for example.

Today, advanced computing methods are applied to problems that occur in older projects, with important consequences for licensing. An example is the utilization of computational simulation methods in fluid dynamics for the analysis of thermal stratification problems verified in PWR reactors. Another recent computational method application is the use of Genetic Algorithms to optimize the nucleus reload.

In the XXIst century, the revived interest and development of new reactor technologies cannot leave aside the best computational technology available.

The computational System

The utilization of clusters of workstations or microcomputers (PCs) has received wide acceptance recently due to the relatively low cost, ease of upgrades, utilization of “open” hardware and software and to its being independent of suppliers and importation licenses.

Our computational system follows the Beowulf-class parallel and distributed computing systems philosophy. The name Beowulf refers to the legendary Anglo-Saxon hero, who, among other great bravery deeds, defeated the terrible monster Grendel.

The research and development of this philosophy of computer cluster organization evolved from the original Beowulf system, composed by 16 DX4 processors. The first Beowulf was installed in 1994 in the Center of Excellence in Space Data and Information Sciences (CESDIS), located at the Goddard Space Flight Center-NASA.

The main Beowulf-class computer cluster characteristics are the utilization of a freeware operational system, such as Linux or FreeBSD, and of a dedicated communication network. In general, the communication of the computational system with the rest of the world is carried out through only one knot (Front End).

Our Beowulf

IEN’s computational system project initially foresees the formation of a 17 PC cluster of 700 MHz Pentium III processors, with 256 Mb RAMs and 10 Gb HDs, adding up to a total of 160 Gb hard disk storage area and 4 Gb RAM. Peak performance will exceed 10 Gflop/s (1 Gflop/s is equivalent to 109 floating point operations per second). The computers will be connected in star topology through a 100 Mb/s Fast Ethernet network. The operational system will be Linux.

The computational programs will be developed using the message-passing paradigm between the system knots. MPI communication libraries with C and Fortran bindings will be used for this purpose.

IEN's Beowulf schematic representation.