[Networkit] measuring memory usage with IPython and the memory_profiler

Christian Staudt christian.staudt at kit.edu
Sat Nov 15 14:53:04 CET 2014


Am 15.11.2014 um 14:07 schrieb Henning Meyerhenke <meyerhenke at kit.edu>:

> I assume the measurements are reasonably correct and refer to what's 
> actually being done in C++, right?

The measurements look sane. One could double-check by comparing to valgrind's output, but I believe valgrind cannot be used with large or parallel computations.

In general, they reflect what happens on the C++ layer and the Python-related overhead should be negligible. I assume that in my code example there is no relevant Python-overhead that would bias experiments on how the memory usage of the algorithm scales. Of course it is possible to measure the wrong thing if one does not pay attention to possible new objects created by the Python code one calls %memit on. 

Christian


-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 495 bytes
Desc: Message signed with OpenPGP using GPGMail
URL: <https://lists.ira.uni-karlsruhe.de/mailman/private/networkit/attachments/20141115/50e1d9f3/attachment.sig>


More information about the NetworKit mailing list