- changed status to open
- removed comment
testsuite AHFinderDirect checkpointML should be removed
The testsuite AHFinderDirect checkpointML should be removed. This is using Exact and there is a corresponding EinsteinExact testsuite. Plus, this testsuite is about 20MG large.
Keyword:
Comments (10)
-
reporter -
- removed comment
I disagree. I am fine with swapping out Exact with EinsteinExact. However this is the only test that test checkpointing. Given that we actually did have failures during checkpointing/recovery in the past (the issues with max_time_levels come to mind) we should not remove all tests that test this functionality. Similarly it should be a test that is somewhat close to a real situation (ie exercise actual production code) rather than a simplified wavetoy code.
Certainly one could try and make the checkpoint file size smaller, though I am not concerned about the 20MiB checkpoint size.
-
reporter - removed comment
checkpointML-EE checks checkpointing as well.
-
- removed comment
Fine by me to remove checkpointML then. Is it any smaller?
-
reporter - removed comment
No, it is of similar size. My comment really only meant that we don't need two of these monsters. As far as I can see they use the same initial data, just two different thorns.
-
- changed status to open
- removed comment
I am fully in agreement with comment:5. One of them is enough. Sorry for the confusion. Should have checked which other tests exist before commenting.
-
- removed comment
Is the checkpoint file compressed?
-
- removed comment
It's not. Compressing with h5repack -f SHUF -f GZIP=9 reduces its size to 11MiB from 19MiB. From the looks of it svn stores it in compressed form on the server though (a local svn repository containing the 19MiB file is only 12MiB large). Subversion also transfers data in compressed form over the network (I think).
So it is uncompressed on the user's disk but compressed everywhere else. We can certainly go ahead and compress the hdf5 file though I don't think it will help speed up up checkouts or reduce load on the server. I will however reduce the checkout size on the user's disk (by ~10MiB since svn keeps two copies around).
-
reporter - removed comment
Yes, we should then compress the EE testsuite checkpoint.
-
reporter - changed status to resolved
- removed comment
- Log in to comment