CW: Cloud computing is also on the agenda at this year's International Supercomputing Conference (ISC '09). Large cloud providers such as Amazon or Google use not so powerful central servers for their IT infrastructure but a large number of standardized x86 computers. Will the classic supercomputers in the cloud age become a phase-out model?
Kranzlmüller: This question can be answered by classifying the applications: High-performance computer applications can be aptly classified in a three-dimensional space whose dimensions are computing power, data volume and communication intensity. The various supercomputing system architectures are usually optimal in one of the three dimensions and offer tradeoffs to the other two dimensions. Clouds provide customizable processing power and possibly large memory with low communication performance. The supercomputers on the top places of the TOP 500, on the other hand, usually offer a special connection solution with high communication capacity. However, the boundaries between these systems are increasingly moving.
CW: In what areas should companies prefer x86 server farms over traditional supercomputers?
Kranzlmüller: In principle, the use of a cloud depends on the application. When it comes to linear processes and independent, relatively smaller amounts of data, then a cloud can be used quite meaningfully. With large memory requirements or closely coupled, parallel processes with correspondingly large I / O requirements, clouds are often unusable compared to supercomputers.