From: Timo Heister <timo.heister@gmail.com>
Date: Wed, 11 May 2022 20:39:40 +0000 (-0400)
Subject: step-40: update IO section
X-Git-Tag: v9.4.0-rc1~183^2~1
X-Git-Url: https://gitweb.dealii.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=8a6c10026a556ea85efaf77f54a82c97fffe8495;p=dealii.git

step-40: update IO section

Update the documentation on how we generate output using MPI I/O (we
originally wrote one vtu per process).
Remove the limit of 32 MPI ranks, because it is confusing according to
my students. I think this was put in place to avoid generating 1000+
output files, which is no longer the case (we use 8 here). The output is
a few MB in total, so I don't see a reason to limit things.
---

diff --git a/examples/step-40/step-40.cc b/examples/step-40/step-40.cc
index de9d4c487e..c97ea27c02 100644
--- a/examples/step-40/step-40.cc
+++ b/examples/step-40/step-40.cc
@@ -545,12 +545,17 @@ namespace Step40
   // bottleneck in parallelizing. In step-18, we have moved this step out of
   // the actual computation but shifted it into a separate program that later
   // combined the output from various processors into a single file. But this
-  // doesn't scale: if the number of processors is large, this may mean that
-  // the step of combining data on a single processor later becomes the
-  // longest running part of the program, or it may produce a file that's so
-  // large that it can't be visualized any more. We here follow a more
-  // sensible approach, namely creating individual files for each MPI process
-  // and leaving it to the visualization program to make sense of that.
+  // doesn't scale for several reasons: First, creating a single file per
+  // processor will overwhelm the filesystem with a large number of processors.
+  // Second, the step of combining all output files later can become the longest
+  // running part of the program, or it may produce a file that's so large that
+  // it can't be visualized easily any more.
+  //
+  // We here follow a more sophisticated approach that uses
+  // high-performance, parallel IO routines using MPI I/O to write to
+  // a small, fixed number of visualization files (here 8). We also
+  // generate a .pvtu record referencing these .vtu files, which can
+  // be opened directly in visualizatin tools like ParaView and VisIt.
   //
   // To start, the top of the function looks like it usually does. In addition
   // to attaching the solution vector (the one that has entries for all locally
@@ -581,7 +586,7 @@ namespace Step40
 
     data_out.build_patches();
 
-    // The next step is to write this data to disk. We write up to 8 VTU files
+    // The final step is to write this data to disk. We write up to 8 VTU files
     // in parallel with the help of MPI-IO. Additionally a PVTU record is
     // generated, which groups the written VTU files.
     data_out.write_vtu_with_pvtu_record(
@@ -595,11 +600,7 @@ namespace Step40
   // The function that controls the overall behavior of the program is again
   // like the one in step-6. The minor difference are the use of
   // <code>pcout</code> instead of <code>std::cout</code> for output to the
-  // console (see also step-17) and that we only generate graphical output if
-  // at most 32 processors are involved. Without this limit, it would be just
-  // too easy for people carelessly running this program without reading it
-  // first to bring down the cluster interconnect and fill any file system
-  // available :-)
+  // console (see also step-17).
   //
   // A functional difference to step-6 is the use of a square domain and that
   // we start with a slightly finer mesh (5 global refinement cycles) -- there
@@ -641,11 +642,10 @@ namespace Step40
         assemble_system();
         solve();
 
-        if (Utilities::MPI::n_mpi_processes(mpi_communicator) <= 32)
-          {
-            TimerOutput::Scope t(computing_timer, "output");
-            output_results(cycle);
-          }
+        {
+          TimerOutput::Scope t(computing_timer, "output");
+          output_results(cycle);
+        }
 
         computing_timer.print_summary();
         computing_timer.reset();