External Memory Sort On CGM1 Clusters

External memory sort has been widely accepted as an overall benchmark to evaluate the processing performance of computers. Lots of algorithms have been developed to sort large scalable data in different environments. The bottlenecks of external memory sort are the I/O operation and communication cost. In this paper, we adapted HPVM MinuteSort[4], and borrowed the THsort[5] idea to develop our external memory sort algorithm, which minimizes the I/O and communication costs. We hope that we will get better performance than other algorithms.