Rob Ross wrote: > > Are you saying that the system didn't have enough free disk space? What > happens in the OpenMPI case in this event? Surely the test shouldn't > complete successfully... > I was testing if the involved filesystem influenced the test results (either the filesystem where the file ends up when writing with the 'default' AD, or the filesystem where the PVFS2 database is on); For this, I also tested on a tmpfs filesystem. I didn't realize it was too small, and only noticed this when I got the 'assertion failed: H5Dwrite succeeded' warnings. I think it was on pvfs2, and I switched the database to another filesystem and restarted the server and continued. At least, the error disappeared, so I assumed it was due to the lack of free space. (Also seems reasonable, H5Dwrite should return an error if a write does not succeed, and for extendible datasets it is likely that you can run out of disk space)
It looked enough like a 'real' error (Assertion (H5Fcreate succeeded) failed) to cause confusion, and that's why I warned about it in my mail. All other tests have been done on filesystems with enough free space. I'm actually less certain that the free space issue was the direct cause of the H5Dwrite warning: * I currently cannot reproduce it on system 3 (not with Open MPI, and also not with mpich2) * If I reduce free space, the big dataset test gets skipped by the test program. It could have been a coincidence (e.g. test starts, and something else puts something in tmp causing a reduction in free space & making the write fail) I cannot access the original system right now, I will have to investigate it later on. For now, I wouldn't assume anything when getting a H5Dwrite error. Greetings, Dries Disclaimer: http://www.kuleuven.be/cwis/email_disclaimer.htm _______________________________________________ Pvfs2-users mailing list [email protected] http://www.beowulf-underground.org/mailman/listinfo/pvfs2-users
