Hi MPB community,
by chance I noticed that the runtime of MPB is not linear with the number
of k-vectors to compute. So I did a small runtime study for different
numbers of k-vectors. I used the 1D bragg example for that, specifying the
number of k-vectors by changing the numberofkvectors in the command
"(interpolate numberofkvectors k-points)". I plotted numberofkvectors vs.
the runtime (from the "total elapsed time for run" at the end of the
out-file) and it can be nicely fit with a quadratic function:
numberofkvectors, runtime [seconds]
5000, 1
10000, 4
20000, 13
50000, 73
100000, 279
200000, 1140
500000, 7297
I would have expected a linear O(n) dependency assuming that each k-point
needs approximately the same effort to compute its solution. Given a
non-linear runtime, it would make sense to split simulations into several
runs with less k-vectors in each simulation...
Am I missing something? Or is this just a special case with that setup?
Thanks in advance!
Johannes
_______________________________________________
mpb-discuss mailing list
mpb-discuss@ab-initio.mit.edu
http://ab-initio.mit.edu/cgi-bin/mailman/listinfo/mpb-discuss
