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.
Do you think writing the data into pvts format will enhance the performance? Can I use some PETSc functions to save the local vector as pvts format and view them as a global vector through paraview/visit? Thank you. Xiangdong On Tue, Mar 4, 2014 at 4:49 PM, Matthew Knepley <[email protected]> wrote: > On Tue, Mar 4, 2014 at 3:43 PM, Xiangdong <[email protected]> wrote: > >> If I use VecView and write the vector into vtk file, is this done in >> parallel or serial? Does each processor send the data to processor 0 and >> let it write to the disk? >> >> Is it possible to let each processor write its own portion and merge them >> together later? >> > > I urge you strongly to avoid premature optimization and worrying. > > If you want to discuss details that is fine, but I do not think the fears > expressed > above are grounded in reality. Network bandwidth is at least as good as > memory > bandwidth, so sending to proc 0 is not a problem below 10,000 procs or so. > You > might think serializing the disk writes would be, but its very likely your > machine > is doing that anyway by having a small number (usually 1) of I/O nodes. > For high > core counts we can use MPI I/O which you can turn on with a command line > argument. > > Matt > > >> Thank you. >> >> Xiangdong >> >> >> On Tue, Mar 4, 2014 at 2:06 PM, Matthew Knepley <[email protected]>wrote: >> >>> On Tue, Mar 4, 2014 at 12:58 PM, Xiangdong <[email protected]> wrote: >>> >>>> Thanks, Matt. >>>> >>>> Given that the FILE_MODE_APPEND is not supported for VTK format, is it >>>> possible to write two vectors into the same VTK data file? >>>> >>> >>> Yes, just call VecView() twice. >>> >>> Matt >>> >>> >>>> Thank you. >>>> >>>> Xiangdong >>>> >>>> >>>> On Tue, Mar 4, 2014 at 1:52 PM, Matthew Knepley <[email protected]>wrote: >>>> >>>>> On Tue, Mar 4, 2014 at 12:31 PM, Xiangdong <[email protected]> wrote: >>>>> >>>>>> What is the difference between PetscViewerSetType and >>>>>> PetscViewerSetFormat? It seems that the first one take argument like >>>>>> PETSCVIEWERVTK, while the second one takes PETSC_VIEWER_VTK_VTS. >>>>>> >>>>> >>>>> A viewer type is the object type, just like other PETSc object, e.g. >>>>> KSP. The format is a particular version >>>>> of that output. For example, ASCII viewer is a type, whereas Matlab is >>>>> a format. >>>>> >>>>> >>>>>> By the way, how can I find a full list of formats? Clearly, >>>>>> PETSC_VIEWER_VTK_VTS is not listed in online documentation for >>>>>> PetscViewerSetFormat >>>>>> >>>>>> http://www.mcs.anl.gov/petsc/petsc-current/docs/manualpages/Viewer/PetscViewerSetFormat.html >>>>>> >>>>> >>>>> It is listed in the complete list: >>>>> >>>>> >>>>> http://www.mcs.anl.gov/petsc/petsc-dev/docs/manualpages/Viewer/PetscViewerFormat.html >>>>> >>>>> Matt >>>>> >>>>> >>>>>> Thank you. >>>>>> >>>>>> Xiangdong >>>>>> >>>>>> >>>>>> On Tue, Mar 4, 2014 at 1:02 PM, Matthew Knepley <[email protected]>wrote: >>>>>> >>>>>>> On Tue, Mar 4, 2014 at 11:44 AM, Xiangdong <[email protected]>wrote: >>>>>>> >>>>>>>> Hello everyone, >>>>>>>> >>>>>>>> When I use PetsViewerVTKOpen to output vec in vtk format, is it in >>>>>>>> ASCII format or binary format? Are there any options to choose between >>>>>>>> them? >>>>>>>> >>>>>>> >>>>>>> It is determined by the format, e.g. PETSC_VIEWER_VTK_VTU. >>>>>>> >>>>>>> Matt >>>>>>> >>>>>>> >>>>>>>> Thank you. >>>>>>>> >>>>>>>> Xiangdong >>>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> -- >>>>>>> What most experimenters take for granted before they begin their >>>>>>> experiments is infinitely more interesting than any results to which >>>>>>> their >>>>>>> experiments lead. >>>>>>> -- Norbert Wiener >>>>>>> >>>>>> >>>>>> >>>>> >>>>> >>>>> -- >>>>> What most experimenters take for granted before they begin their >>>>> experiments is infinitely more interesting than any results to which their >>>>> experiments lead. >>>>> -- Norbert Wiener >>>>> >>>> >>>> >>> >>> >>> -- >>> What most experimenters take for granted before they begin their >>> experiments is infinitely more interesting than any results to which their >>> experiments lead. >>> -- Norbert Wiener >>> >> >> > > > -- > What most experimenters take for granted before they begin their > experiments is infinitely more interesting than any results to which their > experiments lead. > -- Norbert Wiener >
