[Feature-dev] [ feature-Bugs-1311 ] FEATURE results comparison in tests uses too much memory on large files

[noreply at simtk.org]

Bugs item #1311, was opened at 2010-08-15 18:55
You can respond by visiting: 
https://simtk.org/tracker/?func=detail&atid=145&aid=1311&group_id=16

Category: Installation
Group: Memory, bounds, rounding err
Status: Open
Resolution: None
Priority: 3
Submitted By: Teague Sterling (teague)
Assigned to: Nobody (None)
Summary: FEATURE results comparison in tests uses too much memory on large files

Initial Comment:
The Feature_Vector class uses large amounts of memory causing "out of memory" errors on some systems. As this is used heavily in validating results in regession testing, when this error occurs the whole test fails even when the results are in agreement.

On both koomgang and my home system the test for 1GYT --- one of the largest PDB files --- fails randomly on occasion. This appears to be from an out of memory error in perl. When comparing the resulting feature files in diff, they appear identical. This is causing misleading and incorrect results from the regression test.

For large pdb files (approx 50,000+ lines) the feature file diff takes over 4GB of memory. If this is not available the test will crash. This is then interpreted by the test harness as a failure.

Recommending running an initial "coarse" line-by-line diff to check results and only if they are not exaxtly matching should we use feature_vector to check within tolerances. Additionally, feature_vector could free memory of lines as it processes them to reduce its footprint. 

----------------------------------------------------------------------

You can respond by visiting: 
https://simtk.org/tracker/?func=detail&atid=145&aid=1311&group_id=16