Xiangdong <[email protected]> writes: > At this moment, when I write the solutions (two vectors with a few billion > unknowns) to binary vts format, the writing rate is about 100MB/s from proc > 0.
How many procs are you writing from and what filesystem do you have? How much faster is it if you write the PETSc binary format using MPI-IO (-viewer_binary_mpiio)? If that is lots faster for you, I can fairly simply add support for writing VTS that way. You'll still pay when you visualize, however. > Do you think writing the data into pvts format will enhance the > performance? No. PVTS is a crappy format designed by people that evidently did not understand parallel IO performance. But all the VTK formats are crappy if you really care about performance. With the VTK formats, if you write it efficiently, it will still be a bottleneck to read. We support them because they are easy. Use HDF5 if you want something sensible.
pgpx0Z_8LE8_C.pgp
Description: PGP signature
