I did not follow the discussion, but I want to remember (for the case that some are not aware of) that MPI_File_get_info returns not what the user set. It returns all used info, system-specific infos plus those set by the user that are not ignored. This works fine with all the parallel file-systems and their system-specific (i.e., non-user supplied?) values. And of course, the user can learn from that and change those values with MPI_File_set_info. The wording for MPI_Win_get_info and MPI_Comm_get_info is the same. See MPI-3.1, Sect. 6.4.4, page 250, lines 9-13, Advice to users. (Same on page 502, lines 1-5) And other wording in Sect. 11.2.7, page 415-416, lines 47ff, Adv. to impl. Rolf ----- Original Message -----
From: "Daniel Holmes" <[email protected]> To: "MPI WG Point To Point Communications working group" <[email protected]> Sent: Wednesday, March 1, 2017 11:58:28 PM Subject: [mpiwg-p2p] Result of reading of Info Query #63
Exec summary
The MPI_XXX_GET_INFO text is acceptable to everyone as is.
The MPI_XXX_SET_INFO text is contentious for two reasons: 1) (lack of) ability for MPI to pick non-user supplied values. Pavan brought this up. Resolution: none. Suggestions: continue arguing against this or add additional MPI_XXX_SET_INFO_INEXACT (or similar), which allows randomness - lets user choose. 2) (lack of) transactional behaviour - all key/value updates should succeed or all should fail. Martin brought this up. Resolution: this is a new API, we’d love to see your proposal.
Cheers, Dan. -- The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336.
_______________________________________________ mpiwg-p2p mailing list [email protected] https://lists.mpi-forum.org/mailman/listinfo/mpiwg-p2p
-- Dr. Rolf Rabenseifner . . . . . . . . . .. email [email protected] High Performance Computing Center (HLRS) . phone ++49(0)711/685-65530 University of Stuttgart . . . . . . . . .. fax ++49(0)711 / 685-65832 Head of Dpmt Parallel Computing . . . www.hlrs.de/people/rabenseifner Nobelstr. 19, D-70550 Stuttgart, Germany . . . . (Office: Room 1.307)