mvapich-1.0.0 : protocol version 8 not known

[Pete Wyckoff]

louwrens.boonstra at clustervision.com wrote on Fri, 14 Mar 2008 14:03 +0100:
> Hi, I'm getting this error when running mpiexec on mvapich-1.0.0:
>
> mpiexec: Warning: read_ib_one: protocol version 8 not known, but might 
> still work.
> mpiexec: read_ib_one: version 6 startup.
> mpiexec: service_ib_startup: rank 0 in, 7 + 0 left.
> mpiexec: Error: read_ib_one: mixed version executables (6 and 8), no hope.
>
> From earlier messages I understood this has to do with the mvapich version. 
> Any changed this will be fixed soon for 1.0.0 or should I downgrade to 
> mvapich-0.9.8 / 0.9.9 ?

Even though Frank Mietke did the work of figuring out how the new
version 6 worked in mvapich 1.0 beta, it appears they have changed
to version 8 for the release.  Frustrating.  Maybe Frank will also
be interested in looking at this issue.  Help is welcome.

In the mean time I've cheked in a patch to fix the "no hope" bug
above.  It should keep trying even if it doesn't know version 8.
There is practically no chance that this will solve your problem,
so please keep using 0.9.9 for a while.

		-- Pete


Modified: trunk/ib.c
===================================================================
--- trunk/ib.c	2008-03-11 18:57:43 UTC (rev 414)
+++ trunk/ib.c	2008-03-14 14:38:02 UTC (rev 415)
@@ -26,6 +26,7 @@
 
 /* parameters for initial data read from each process */
 static int version = -1;
+static int version_as_read = -1;  /* to check for mixed version only */
 static char *address = 0;
 static int address_size = 0;
 static char *pids = 0;
@@ -251,19 +252,19 @@
 
     if (version == -1) {
 	version = testvers;
+	version_as_read = testvers;
 	if (!(version == 1 || version == 2 || version == 3 || version == 5 ||
 	      version == 6)) {
-	    warning(
-	      "%s: protocol version %d not known, but might still work",
-	      __func__, version);
+	    warning("%s: protocol version %d not known, but might still work",
+	            __func__, version);
 	    version = 6;  /* guess the latest still works */
 	}
 	debug(1, "%s: version %d startup%s", __func__, version,
-	  non_versioned_092 ? " (unversioned)" : "");
+	      non_versioned_092 ? " (unversioned)" : "");
     } else {
-	if (version != testvers)
+	if (version_as_read != testvers)
 	    error("%s: mixed version executables (%d and %d), no hope",
-	      __func__, version, testvers);
+	          __func__, version_as_read, testvers);
     }
     if (rank < 0 || rank >= numtasks)
 	error("%s: rank %d out of bounds [0..%d)", __func__, rank, numtasks);