Artificial intelligence

Abstract: High performance cluster computers are very popular due to its low price and scalability. But several new issues must be attended while programming on such systems. The most important are bandwidth of the interconnect and I/O storage speed. Problem of uniform data processing can be described in terms of queueing theory for estimations of computation efficiency. For several parallel algorithms theoretical estimations are already known. In this article we propose an universal experimental approach based on computer simulation. Using this approach we can evaluate many properties of computation process for a specified program on a specified cluster architecture.

Keywords:

References: