COB-5: Lower MPI polling as it seems to be good enough and stresses the system less.

RTCP/Cobalt/InputProc/src/Transpose/MPIUtil.cc

@@ -446,11 +446,13 @@ namespace LOFAR {
       ScopedLock sl(mutex);
 
       while(!done) {
-        // next poll will be in 0.1 ms
+        // next poll will be in 0.5 ms
         //
         // NOTE: MPI is VERY sensitive to this, requiring
         //       often enough polling to keep transfers
         //       running smoothly.
+        //
+        // If the observations do not run at real time, try lowering the polling interval.
 
         if (requests.empty()) {
           // wait for request, with lock released
@@ -463,7 +465,7 @@ namespace LOFAR {
           // the lock and just wait with a timeout
           if (!requests.empty()) {
             struct timespec deadline = TimeSpec::now();
-            TimeSpec::inc(deadline, 0.0001);
+            TimeSpec::inc(deadline, 0.0005);
 
             newRequest.wait(mutex, deadline);
           }