From: Vermaas, Joshua (Joshua.Vermaas_at_nrel.gov)
Date: Sun Apr 09 2017 - 20:58:47 CDT
Hi Jim,
outputTiming gives an estimate for when the simulation will be done, and is separate from benchmark lines. This is what a sample line would look like:
TIMING: 50900 CPU: 16780.1, 0.333541/step Wall: 16805.9, 0.333584/step, 0.00926621 hours remaining, 766.042969 MB of memory in use.
Benchmark lines are put out only at the beginning of the simulation as the work units get moved around the different processors. Sample lines would look like:
Info: Benchmark time: 10 CPUs 0.325354 s/step 1.88284 days/ns 749.938 MB memory
I tend to like using the benchmark lines for picking the best running configuration for the hardware at my disposal, since the first line comes out at 400 steps or so, and nicely puts it in days/ns without me needing to deal with the conversion myself. I’ve found timing most useful when I’m trying to estimate a walltime on a queuing system that lets me adjust the run times after the fact.
Josh Vermaas
Director’s Postdoctoral Fellow
National Renewable Energy Laboratory
joshua.vermaas_at_nrel.gov<mailto:joshua.vermaas_at_nrel.gov>
On Apr 7, 2017, at 4:31 PM, Jim <jim.jim.strong_at_gmail.com<mailto:jim.jim.strong_at_gmail.com>> wrote:
Hello,
The standard output option
outputTiming
specifies the timing output to stdout ("TIMING:").
However, what I don't understand is:
* what option in the config file specifies the Benchmark output ("Info: Benchmark time:")
* how are the two outputs different
* when benchmarking, should one use the "Benchmark" or the "Timing" output, or both, or instead calculate timing metrics from the "WallClock/CPUTime" outputs from the end of the program?
Thanks.
This archive was generated by hypermail 2.1.6 : Mon Dec 31 2018 - 23:20:12 CST