diff -ruN pbs-2.2.11-stock/README.pw pbs-2.2.11/README.pw
--- pbs-2.2.11-stock/README.pw	Wed Dec 31 19:00:00 1969
+++ pbs-2.2.11/README.pw	Fri Dec 22 12:13:08 2000
@@ -0,0 +1,35 @@
+#!/bin/bash
+#
+# Call configure with the right stuff used locally.
+#
+./configure \
+  --prefix=/usr/local/pbs-2.2.11 \
+  --set-server-home=/var/spool/pbs
+# now do:  make -j3 MAKE="make -j3"
+
+#
+# Documentation of changes applied as part of the mpiexec patch.
+#
+##
+# src/include/tm.h
+# Fixed misplaced "typedef" in struct tm_whattodo.h:89.
+#
+# src/lib/Libifl/rpp.c
+# Avoid EINTR in recvfrom by looping it, else bogus error result.
+# Incrcease rpp retry count.
+#
+# src/lib/Libifl/tm.c
+# Do not call DIS_tcp_setup(-1) if connection to the local mom failed
+# after 5 tries.
+#
+# src/resmom/mom_comm.c
+# Fix off-by-one error in saving of envp during TM_SPAWN processing.  Avoid
+# some coredumps due to corruption elsewhere.
+#
+# src/resmom/start_exec.c
+# Added extension to get stdin/out/err files from the environment rather than
+# always using /dev/null during tm_spawn.
+#
+# src/include/net_connect.h
+# Increase max connection idle time.
+#
diff -ruN pbs-2.2.11-stock/src/include/net_connect.h pbs-2.2.11/src/include/net_connect.h
--- pbs-2.2.11-stock/src/include/net_connect.h	Mon Oct 18 17:22:01 1999
+++ pbs-2.2.11/src/include/net_connect.h	Mon Dec 18 15:44:12 2000
@@ -62,7 +62,8 @@
 #define PBS_NET_TYPE
 #endif
 
-#define PBS_NET_MAXCONNECTIDLE  900
+/* for use with mpiexec, connection may last quite a bit longer than 15 min */
+#define PBS_NET_MAXCONNECTIDLE  2592000
 #define PBS_NET_CONN_AUTHENTICATED 1
 #define PBS_NET_CONN_FROM_PRIVIL  2
 
diff -ruN pbs-2.2.11-stock/src/include/tm.h pbs-2.2.11/src/include/tm.h
--- pbs-2.2.11-stock/src/include/tm.h	Wed Jul 21 18:56:43 1999
+++ pbs-2.2.11/src/include/tm.h	Mon Dec 18 15:44:12 2000
@@ -82,11 +82,11 @@
 **	The tm_whattodo structure contains data for the last
 **	tm_register event polled.  This is not implemented yet.
 */
-struct	tm_whattodo {
+typedef struct	tm_whattodo {
 	int		tm_todo;
 	tm_task_id	tm_what;
 	tm_node_id	tm_where;
-}	typedef		tm_whattodo_t;
+}	tm_whattodo_t;
 
 /*
 **	Prototypes for all the TM API calls.
diff -ruN pbs-2.2.11-stock/src/lib/Libifl/rpp.c pbs-2.2.11/src/lib/Libifl/rpp.c
--- pbs-2.2.11-stock/src/lib/Libifl/rpp.c	Fri Jul 23 16:58:57 1999
+++ pbs-2.2.11/src/lib/Libifl/rpp.c	Mon Dec 18 15:44:12 2000
@@ -148,7 +148,7 @@
 			         /*every RPP_TIMEOUT seconds              */
 
 
-#define	RPP_RETRY	20       /*after this many sendto attempts on a   */
+#define	RPP_RETRY	200       /*after this many sendto attempts on a   */
 			         /*packet, give up and declare the stream */
                                  /*to be in the "stale" state             */
 
@@ -1154,8 +1154,13 @@
 	assert(data != NULL);
 
 	flen = sizeof(struct sockaddr_in);
-	len = recvfrom(fd, data, RPP_PKT_SIZE, 0,
-			(struct sockaddr *)&addr, &flen);
+	/*
+	 * loop added to avoid hanging up due to receipt of a signal  --pw
+	 */
+	do {
+		len = recvfrom(fd, data, RPP_PKT_SIZE, 0,
+				(struct sockaddr *)&addr, &flen);
+	} while (len == -1 && errno == EINTR);
 	if (len == -1) {
 		free(data);
 		if (errno == EWOULDBLOCK ||
diff -ruN pbs-2.2.11-stock/src/lib/Libifl/tm.c pbs-2.2.11/src/lib/Libifl/tm.c
--- pbs-2.2.11-stock/src/lib/Libifl/tm.c	Fri Apr 16 15:08:19 1999
+++ pbs-2.2.11/src/lib/Libifl/tm.c	Mon Dec 18 15:44:12 2000
@@ -411,7 +411,9 @@
 		}
 
 	}
-	DIS_tcp_setup(local_conn);
+	if (local_conn >= 0)
+	    DIS_tcp_setup(local_conn);
+
 	return (local_conn);
 }
 
diff -ruN pbs-2.2.11-stock/src/resmom/mom_comm.c pbs-2.2.11/src/resmom/mom_comm.c
--- pbs-2.2.11-stock/src/resmom/mom_comm.c	Thu Feb 17 17:56:51 2000
+++ pbs-2.2.11/src/resmom/mom_comm.c	Mon Dec 18 15:44:12 2000
@@ -537,6 +537,19 @@
 		stream_addr = rpp_getaddr(stream);
 		node_addr = rpp_getaddr(hp->hn_stream);
 
+		/*
+		 * these might become corrupt, avoid coredump  --pw
+		 */
+		if (!stream_addr) {
+		    sprintf(log_buffer, "find_node: stream id %d does not exist",
+		      stream);
+		    log_err(-1, id, log_buffer);
+		}
+		if (!node_addr) {
+		    sprintf(log_buffer, "find_node: node id %d does not exist",
+		      hp->hn_stream);
+		    log_err(-1, id, log_buffer);
+		}
 		if (memcmp(&stream_addr->sin_addr, &node_addr->sin_addr,
 				sizeof(node_addr->sin_addr)) != 0) {
 			sprintf(log_buffer,
@@ -2581,7 +2594,10 @@
 				free(env);
 				break;
 			}
-			if (i == numele) {
+			/* need to remember extra slot for NULL at
+			 * the end  --pw
+			 */
+			if (i == numele-1) {
 				numele *= 2;
 				envp = (char **)realloc(envp,
 					numele*sizeof(char **));
diff -ruN pbs-2.2.11-stock/src/resmom/start_exec.c pbs-2.2.11/src/resmom/start_exec.c
--- pbs-2.2.11-stock/src/resmom/start_exec.c	Thu Feb 17 17:56:52 2000
+++ pbs-2.2.11/src/resmom/start_exec.c	Fri Dec 22 12:12:05 2000
@@ -1360,6 +1360,42 @@
 }
 
 /*
+ * Look for a certain environment variable which has a port# which should
+ * be opened on the MS to establish communication for one of the 3 stdio
+ * streams.  >=0 return is that valid fd, -1 means no env var found,
+ * -2 means malformed env value or failure to connect.
+ */
+static int
+search_env_and_open(const char *envname, u_long ipaddr)
+{
+    static char *id = "search_env_and_open";
+    int i, len = strlen(envname);
+
+    for (i=0; i<vtable.v_used; i++)
+	if (!strncmp(vtable.v_envp[i], envname, len)) {
+	    const char *cp = vtable.v_envp[i] + len;
+	    char *cq;
+	    int fd, port;
+	    if (*cp++ != '=') break;  /* empty, ignore it */
+	    port = strtol(cp, &cq, 10);
+	    if (*cq) {
+		log_err(errno, id, "improper value for MPIEXEC_STD*_PORT");
+		return -2;
+	    }
+#if 0       /* debugging */
+	    log_err(0, "search_env_and_open attempting open", vtable.v_envp[i]);
+#endif
+	    if ((fd = open_demux(ipaddr, port)) < 0) {
+		log_err(errno, id, "failed connect to mpiexec process on MS");
+		log_err(errno, id, vtable.v_envp[i]);
+		return -2;
+	    }
+	    return fd;
+	}
+    return -1;  /* not found */
+}
+
+/*
 ** Start a process for a spawn request.  This will be different from
 ** a job's initial shell task in that the environment will be specified
 ** and no interactive code need be included.
@@ -1377,7 +1413,7 @@
 	int	pipes[2], kid_read, kid_write, parent_read, parent_write;
 	int	pts;
 	int	i, j;
-	int	fd;
+	int	fd0, fd1, fd2;
 	u_long	ipaddr;
 	struct	array_strings	*vstrs;
 	struct  startjob_rtn sjr;
@@ -1598,30 +1634,40 @@
 	/*
 	** Set up stdin.
 	*/
-	if ((fd = open("/dev/null", O_RDONLY)) == -1) {
+	/* look through env for a port# on MS we should use for stdin */
+	if ((fd0 = search_env_and_open("MPIEXEC_STDIN_PORT", ipaddr)) == -2)
+	    starter_return(kid_write, kid_read, JOB_EXEC_FAIL2, &sjr);
+	/* use /dev/null if no env var found */
+	if (fd0 < 0 && (fd0 = open("/dev/null", O_RDONLY)) == -1) {
 		log_err(errno, "newtask", "could not open devnull");
 		(void)close(0);
 	}
 	else {
-		(void)dup2(fd, 0);
-		if (fd > 0)
-			(void)close(fd);
+		(void)dup2(fd0, 0);
+		if (fd0 > 0)
+			(void)close(fd0);
 	}
 
+	/* look through env for a port# on MS we should use for stdout/err */
+	if ((fd1 = search_env_and_open("MPIEXEC_STDOUT_PORT", ipaddr)) == -2)
+	    starter_return(kid_write, kid_read, JOB_EXEC_FAIL2, &sjr);
+	if ((fd2 = search_env_and_open("MPIEXEC_STDERR_PORT", ipaddr)) == -2)
+	    starter_return(kid_write, kid_read, JOB_EXEC_FAIL2, &sjr);
+
 	if (pjob->ji_numnodes > 1) {
 		/*
 		** Open sockets to demux proc for stdout and stderr.
 		*/
-		if ((fd = open_demux(ipaddr, pjob->ji_stdout)) == -1)
+		if (fd1 < 0 && (fd1 = open_demux(ipaddr,pjob->ji_stdout)) == -1)
 		      starter_return(kid_write, kid_read, JOB_EXEC_FAIL2, &sjr);
-		(void)dup2(fd, 1);
-		if (fd > 1)
-			(void)close(fd);
-		if ((fd = open_demux(ipaddr, pjob->ji_stderr)) == -1)
+		(void)dup2(fd1, 1);
+		if (fd1 > 1)
+			(void)close(fd1);
+		if (fd2 < 0 && (fd2 = open_demux(ipaddr,pjob->ji_stderr)) == -1)
 		      starter_return(kid_write, kid_read, JOB_EXEC_FAIL2, &sjr);
-		(void)dup2(fd, 2);
-		if (fd > 2)
-			(void)close(fd);
+		(void)dup2(fd2, 2);
+		if (fd2 > 2)
+			(void)close(fd2);
 	
 		(void)write(1,pjob->ji_wattr[(int)JOB_ATR_Cookie].at_val.at_str,
 		     strlen(pjob->ji_wattr[(int)JOB_ATR_Cookie].at_val.at_str));
@@ -1631,17 +1677,41 @@
 
             (pjob->ji_wattr[(int)JOB_ATR_interactive].at_val.at_long > 0)) {
 		/* interactive job, single node, write to pty */
-		if ((pts = open_pty(pjob)) < 0) {
+		pts = -1;
+		if (fd1 < 0 || fd2 < 0) {
+		    if ((pts = open_pty(pjob)) < 0) {
 			log_err(errno, id,"cannot open slave");
 			starter_return(kid_write, kid_read,JOB_EXEC_FAIL1,&sjr);
+		    }
+		    if (fd1 < 0)
+			fd1 = pts;
+		    if (fd2 < 0)
+			fd2 = pts;
 		}
-		(void)dup2(pts, 1);
-                (void)dup2(pts, 2);
-
+		(void)dup2(fd1, 1);
+		(void)dup2(fd2, 2);
+		if (fd1 != pts)
+		    (void) close(fd1);
+		if (fd2 != pts)
+		    (void) close(fd2);
 	} else {
 		/* normal batch job, single node, write straight to files */
-		if (open_std_out_err(pjob) == -1) {
+		pts = -1;
+		if (fd1 < 0 || fd2 < 0) {
+		    if (open_std_out_err(pjob) == -1)
 			starter_return(kid_write, kid_read,JOB_EXEC_FAIL1,&sjr);
+		}
+		if (fd1 >= 0) {
+		    (void) close(1);
+		    (void) dup2(fd1, 1);
+		    if (fd1 > 1)
+			(void) close(fd1);
+		}
+		if (fd2 >= 0) {
+		    (void) close(2);
+		    (void) dup2(fd2, 2);
+		    if (fd2 > 2)
+			(void) close(fd2);
 		}
 	}