Fig. 4.

Download original image
Comparison of the fraction of runtime spent on I/O and on compute for SEG-Y files on a Lustre filesystem and on a Lustre-backed IME filesystem for a range of file sizes for the ExSeisFlow sort benchmark run on 32 cores. The execution time for the Lustre and Lustre + IME runs are individually scaled to unity, and the read, write, and compute time displayed as a fraction of the overall time. While the sort is I/O bound on the Lustre system, it is compute bound on the Lustre + IME system.