Scheduler demonstrator for a parallel processor based computer cluster: STFC Global Challenges Concepts Fund

点击翻译按钮获取中文摘要

There are increasingly large data sets available to researchers, both created by dedicated experiments and collected fromthe internet as web and twitter textual archives. In both cases the rate of increase of data production is growing faster thanthe resources to analyse them with standard CPU based computers. New computer models based on GPUs that weredeveloped for gaming primarily by NVidia and Xeon-phi systems from Intel add co-processor capabilities to computers thatallow much more rapid data processing using the hundreds of dedicated processors on each board.Using the new computing models provided by these hardware developments means providing a way to use the computerswith these co-processor cards in clusters. By demonstrating the code changes required and requirements for host and coprocessorresources allows a cluster scheduler to use these computers as part of a mixed cluster. With this scheduler inplace it will be simple to extend existing clusters to use predominately co-processor based systems if the problem is onethat they do accelerate. This will allow scientists and businesses looking to process very large data sets to optimise thecosts of the computing resources they must buy to achieve this.