OpenMP Timing Comparisons
To have some idea of how the Katana performs, shown below is the timing comparison of a Boundary Element
Method code parallelized with OpenMP directives. This code consists of two major
sections: 1) compute two N x N matrices whose elements are surface integrals
in parallel; 2) Solving a linear
system Ax = By for x. The bar graph below reports only the timings of the
computation of the matrices.
Note that the result shown below is by no means indicative of a
general performance trend of the Katana Cluster as a code's performance
is highly dependent on the hardware, OS, and compiler.
Figure 1. OpenMP timing comparison of the IBM pSeries versus the Katana.
|
