pascal.deveze at bull.net
2009-Jun-05 11:21 UTC
[Lustre-discuss] Réf. : Re: [ROMIO Req #940] a new Lustre ADIO driver]
Hi Rob, I''ve compiled the romio coming with mpich2 version 1.1rc1 and added some modifications, I ran the script ''runtests'' and I still have problems with 4 tests: i_noncontig : abort noncontig: abort noncontigcoll: errors noncontigcoll2: Fatal errors All other tests are OK. I''ll take you informed of my progress. Pascal |---------+---------------------------> | | Rob Latham | | | <robl at mcs.anl.go| | | v> | | | | | | 03/06/2009 23:36| | | | |---------+---------------------------> >------------------------------------------------------------------------------------------------------------------------------| | | | Pour : pascal.deveze at bull.net | | cc : lustre-discuss at lists.lustre.org | | Objet : Re: [Lustre-discuss] [ROMIO Req #940] a new Lustre ADIO driver] | >------------------------------------------------------------------------------------------------------------------------------| On Wed, May 27, 2009 at 10:15:29AM +0200, pascal.deveze at bull.net wrote:> Without these changes, len_list[i] (where i is 1,3,5,7 ...) in theroutine> ADIOI_LUSTRE_Calc_my_req > are null. > > I will now pass some tests and I hope make contributions.Hi Pascal. Did these tests fix things for you? You''re right: one of the bigger changes from 1.0 to 1.1 was an effort to make ROMIO more 64-bit clean. If you go to the ''romio/test'' directory, there''s a script ''runtests'' that will run through a good number of MPI-IO tests: ''runtests -fname=lustre:/path/to/lustre'' If you run that, let me know how it goes. ==rob -- Rob Latham Mathematics and Computer Science Division Argonne National Lab, IL USA
Rob Latham
2009-Jun-05 14:47 UTC
[Lustre-discuss] Réf. : Re: [ROMIO Req #940] a new Lustre ADIO driver]
On Fri, Jun 05, 2009 at 01:21:28PM +0200, pascal.deveze at bull.net wrote:> I ran the script ''runtests'' and I still have problems with 4 tests: > i_noncontig : abort > noncontig: abort > noncontigcoll: errors > noncontigcoll2: Fatal errors > > All other tests are OK. > > I''ll take you informed of my progress.i_noncontig and noncontig require fcntl() locks, which Lustre supports only if you mount with a special mount option (I don''t remember what that is). Was that the cause for the ''abort'' (it should be pretty clear from the error messages). If coll_test passed, that''s very good progress. Can you tell me more about the noncontig_coll and noncontig_coll2 test failures? thanks ==rob -- Rob Latham Mathematics and Computer Science Division Argonne National Lab, IL USA