<meta http-equiv="Content-Type" content="text/html; charset=utf-8"><div dir="ltr"><div>MPI_File_read_shared / MPI_File_write_shared on NFS seems to be
very slow when running on multi-node environment compared to a single
node run. originally detected with the mpich test case (<b>writeshf90</b>)<br><br>Here is a minimal reproducer with MPI_File_read_shared,<br><a href="https://gist.github.com/AboorvaDevarajan/db4797e02ac71f60384a8254ae2862c8" target="_blank">https://gist.github.com/<wbr>AboorvaDevarajan/<wbr>db4797e02ac71f60384a8254ae2862<wbr>c8</a><br><br><br>[aboorvad@c712f6n06 io]$ cat hostfile_mpich<br>c712f6n06:2<br>c712f6n07:2<br>[aboorvad@c712f6n06 io]$ time mpirun -np 4 --hostfile hostfile_mpich ./read_shared<br><b>real 0m14.709s</b><br>user 0m27.703s<br>sys 0m1.438s<br><br>[aboorvad@c712f6n06 io]$ time mpirun -np 4 ./read_shared<br><br><b>real 0m0.247s</b><br>user 0m0.153s<br>sys 0m0.616s<br><br>Is it an expected behaviour? If yes, why? <br></div><br><br>Thanks.</div>