** Next:** 3D Lid-driven Cavity Flow
** Up:** Methodology
** Previous:** Pre-processing
** Contents**

###

Timing and Performance

For each case, we present the timing and performance results in
both a table and graphically in a log-log graph.

The tables all contain the wall clock time, in seconds, for that run.
Also, for each time, the `Performance' is also presented, where
Performance figures are calculated simply as the time on *n*/2 cores
divided by the time on *n* cores. Thus, a code which scales perfectly,
i.e. if we were to double the cores, then we would half the execution
time, the associated performance value will be 2.0. Any run with an
assoicated performance figure of less than 1.4 has an efficiency less
than 71% and, thus, is not considered to scale. Therefore, we may determine the
optimum number of cores for a given simulation such that it is the largest
number of cores where the Performance is greater than 1.4.

The figures are all plotted on a log-log graph, where each run is
plotted in conjuncture with its associated perfect scaling curve.

Lastly, for all cases, the value of `writeInterval` in `
controlDict` is set to be larger than total number of iterations. This
ensures that no simulation snapshots are written, thereby reducing the
output to be just the standard output.

** Next:** 3D Lid-driven Cavity Flow
** Up:** Methodology
** Previous:** Pre-processing
** Contents**
*Gavin J Pringle*

*2010-04-16*