Rockets, reactors, particle accelerators and supercomputer have one thing in common. If it does not work with force, use more to reach your goal.
Writing applications that run on multiple processors is a short exercise for a developer. Writing a scalable application which runs on ten thousands of processors is a challenging task. Each order of magnitude reveals new insights in the implementation of the application.
In order to program a scalable application, it is crucial to understand the differences between workstations, high performance clusters and supercomputers. As an example, communication between computation nodes is a lesser performance bottleneck for supercomputers than for connected workstations. Nevertheless many performance issues are related to communication.
Beside the architecture of the hardware the type of algorithm and its implementation determine the performance. The decomposition of a regular grid will show why a high body mass index is beneficial for penguins at the South Pole.
This seminar talk will give a brief overview of what a supercomputer is and things to consider for a scalable application.
Description