Back to the main page.

Bug 1537 - memtic/memtoc do not always provide accurate results, leading to jobs getting killed

Reported 2012-06-19 11:07:00 +0200
Modified 2012-08-23 10:35:16 +0200
Product: FieldTrip
Component: peer
Version: unspecified
Hardware: PC
Operating System: Windows
Importance: P3 normal
Assigned to: Eelke Spaak
Depends on:
Blocks: 1041
See also:

Eelke Spaak - 2012-06-19 11:07:42 +0200

For istance, I ran something like this: memtic; ... = ft_preprocessing(...); ... = ft_preprocessing(...); ... = ft_preprocessing(...); memtoc; and it gave me an output of some 300 MB. However, when I submitted the exact same code as a job to qsubexec, with a memreq of 1e9 (so quite a margin even!) the job was killed. When I substantially increased the memreq to 4e9, the job succeeded, and then indeed the feedback returned was something like "Job XXX returned, it required XXX seconds and 3.4 GB".

Eelke Spaak - 2012-06-19 11:08:33 +0200

(In reply to comment #0) Sorry, of course I meant 'submitted the exact same code as a job to *qsubcellfun*'.

Jan-Mathijs Schoffelen - 2012-06-27 13:38:04 +0200

discussed in FT-meeting: it is a consequence of how matlab deals with the memory usage. The memtic/memtoc can either over or under estimate the useage. Only when starting from a totally fresh session, it will actually reflect the true memory requirements. Since this appears to be a matlab feature it cannot be really dealt with in the code. Outcome: rephrasing the statement that is printed regarding memory useage should be adjusted. Byproduct: remove the memory statement for windows.

Eelke Spaak - 2012-07-26 11:44:49 +0200

bash-3.2$ svn commit Sending utilities/memtoc.m Sending utilities/private/ft_postamble_callinfo.m Transmitting file data .. Committed revision 6292.

Eelke Spaak - 2012-08-23 10:35:16 +0200

closing my bugs