An improvement of I/O function for auxiliary storage
Umpei Nagashima, Fumio Nishimoto, Takashi Shibata, Hiroshi Itoh, Minoru Gotoh · ACM SIGARCH Computer Architecture News · 1990
New I/O technique for external auxiliary storage: magnetic disk unit, has been developed to improve the I/O performance on HITAC VOS3/ES1 with usual hardware architecture. Since the I/O technique is based on the idea that the sequence of I/O processes should be divided to some groups and be executed in parallel, and is quite similar to a pipeline on vector computer, we call it Parallel I/O (PIO). The PIO function has been realized and opened to end-users without any change to previous protocols. Comparison of actual performance of PIO with 16 degrees of parallelization (16 PIO), the speed of the PIO is about 26 times faster than the standard (sequential) I/O technique by using chained schedule function. In the case of 8 PIO, PIO is 13 times faster than the standard I/O technique. The efficiency of PIO correlates well with the degree of parallelization. Elapsed time is inversely proportional to the degree of parallelization. The change of acceleration ratio of PIO to the standard I/O is normally proportional to the degree of parallelization. The PIO is quite efficient for large scale supercomputing because I/O between external auxiliary storage and CPU is indispensable in large scale computations. Since the performance of I/O has not been improved, while the performance of supercomputer has made remarkable progress in CPU speed, I/O speed may become bottleneck for computations with supercomputers.