Parallel External Sort of Floating-Point Data by Integer Conversion

This paper introduces a fast external sorting algorithm of floating point numbers with integer operations only, which shortens the computing time significantly. Conversion overhead to integer can be avoided if the floating point data are stored in the disk since integer conversion is made while they are read from the data file. Experimentally large-scale data stored in the disk are sorted in a cluster computer with various data distributions, where speedups over two fold or more are observed. Key-Words: sort, floatingpoint arithmetic, NOWsort, cluster computer, load balancing