MPICH2 Group: During mvapich2 testing we ran into a problem with the NAS benchmark. We've been able to duplicate this issue with mpich2-1.0.7rc2. I built mpich2 with the default settings (./configure && ./make). We executed the test using 4 processes across two nodes. We set the class to B. The test fails due to a collective abort early in the benchmark and generates 4 core files. Each core file presented the same backtrace as below... [perkinjo@ro0-cr NPB3.2-MPI]$ gdb -batch bin/bt.B.4.mpi_io_full core.9445 -x gdb.cmd (no debugging symbols found) Using host libthread_db library "/lib64/tls/libthread_db.so.1". warning: core file may not match specified executable file. Core was generated by `./bin/bt.B.4.mpi_io_full'. Program terminated with signal 11, Segmentation fault. (no debugging symbols found) (no debugging symbols found) (no debugging symbols found) (no debugging symbols found) (no debugging symbols found) (no debugging symbols found) (no debugging symbols found) #0 0x0000000000437dbb in ADIOI_Calc_my_req () #0 0x0000000000437dbb in ADIOI_Calc_my_req () #1 0x00000000004358ef in ADIOI_GEN_WriteStridedColl () #2 0x00000000004634c1 in MPIOI_File_write_all () #3 0x000000000042f474 in PMPI_File_write_at_all () #4 0x0000000000417c5c in pmpi_file_write_at_all__ () #5 0x00000000004171d9 in output_timestep__ () #6 0x0000000000402e47 in MAIN__ () #7 0x0000000000481a42 in main () Have you guys seen this problem before? -- Jonathan Perkins http://www.cse.ohio-state.edu/~perkinjo
There is a known problem caused by a bug in ROMIO's info handling. ROMIO expects some info keys to be passed in MPI_File_open. If it's the same problem, try replacing MPI_INFO_NULL with info in the MPI_File_open in BT/full_mpiio.f (around line 139). Rajeev
-----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Jonathan L. Perkins Sent: Thursday, April 03, 2008 3:49 PM To: [email protected] Cc: [email protected] Subject: [mpich2-core] NAS BTIO Full Failure
MPICH2 Group: During mvapich2 testing we ran into a problem with the NAS benchmark. We've been able to duplicate this issue with mpich2-1.0.7rc2. I built mpich2 with the default settings (./configure && ./make).
We executed the test using 4 processes across two nodes. We set the class to B. The test fails due to a collective abort early in the benchmark and generates 4 core files. Each core file presented the same backtrace as below...
[perkinjo@ro0-cr NPB3.2-MPI]$ gdb -batch bin/bt.B.4.mpi_io_full core.9445 -x gdb.cmd (no debugging symbols found) Using host libthread_db library "/lib64/tls/libthread_db.so.1".
warning: core file may not match specified executable file. Core was generated by `./bin/bt.B.4.mpi_io_full'. Program terminated with signal 11, Segmentation fault. (no debugging symbols found) (no debugging symbols found) (no debugging symbols found) (no debugging symbols found) (no debugging symbols found) (no debugging symbols found) (no debugging symbols found) #0 0x0000000000437dbb in ADIOI_Calc_my_req () #0 0x0000000000437dbb in ADIOI_Calc_my_req () #1 0x00000000004358ef in ADIOI_GEN_WriteStridedColl () #2 0x00000000004634c1 in MPIOI_File_write_all () #3 0x000000000042f474 in PMPI_File_write_at_all () #4 0x0000000000417c5c in pmpi_file_write_at_all__ () #5 0x00000000004171d9 in output_timestep__ () #6 0x0000000000402e47 in MAIN__ () #7 0x0000000000481a42 in main ()
Have you guys seen this problem before?
-- Jonathan Perkins http://www.cse.ohio-state.edu/~perkinjo
participants (2)
-
Jonathan L. Perkins -
Rajeev Thakur