Roland
Thorn Carpet does not yet count communicated bytes; this feature was never completed. However, thorn CarpetLib has timers that do count communicated bytes:
There are two similar timing classes, one defined in Carpet/Timers's "Timer.hh", the other in CarpetLib's "timestat.hh". Unfortunately, both are called "Timer", although they live in different namespaces (Timers and CarpetLib, respectively). This is a result of an incomplete transition from timestat.hh to the Timers thorn. Only timestat.hh counts bytes as well.
(Incidentally, I am currently transitioning to a third timing class which is (a) thread-safe (thorn Timers isn't) and (b) uses C++ features instead of thorn CycleClock.)
To find out details, look at the routine "CarpetLib_printtimestats" that outputs the timing statistics with the byte counts. This is rather low-level output, and is not aggregated across processes. There are a few CarpetLib parameters with "timestat" in their name that control this. By default, these files are called "carpetlib-timing-statistics".
These statistics contain actual byte counts, gathered next to the actual MPI statements. Thorn Carpet might also produce higher-level estimates based on numbers of grid functions, component sizes, numbers of ghost zones etc.; these numbers would be less reliable.
-erik
On Tue, Feb 14, 2017 at 12:38 AM, Roland Haas rhaas@illinois.edu wrote:
Hello Erik, all,
is there a (simple?) way to output the total amount of data communicated in a run? I tried Carpet::timing and its comm_bytes_count however that one is only set in Carpet/src/Timing.c's EndTimingCommunication function which does not seem to be called anywhere anymore.
Yours, Roland
-- My email is as private as my paper mail. I therefore support encrypting and signing email messages. Get my PGP key from http://keys.gnupg.net.
Users mailing list Users@einsteintoolkit.org http://lists.einsteintoolkit.org/mailman/listinfo/users