LAUSR

This paper describes a scheduling method focusing on exploiting the local access of a nonuniform storage-access file system. In our approach, the file access cost is calculated and combined with the CPU load average into a comprehensive value, which will be used as the standard for scheduling. We evaluated our approach in comparison with the original Torque scheduler using three benchmarks: thput-gfpio, readgf, and BLAST benchmarks. For thput-gfpio, the read throughput showed a 3.59 times boost, whereas for readgf, the total execution time was reduced to about 1/10th of the original value. Finally, using the BLAST benchmark, the total execution time was reduced by 33%.