diff --git a/ompi/mca/coll/base/coll_base_alltoallv.c b/ompi/mca/coll/base/coll_base_alltoallv.c
index efe054d696f..d9f821ed28d 100644
--- a/ompi/mca/coll/base/coll_base_alltoallv.c
+++ b/ompi/mca/coll/base/coll_base_alltoallv.c
@@ -14,7 +14,7 @@
  * Copyright (c) 2013      Los Alamos National Security, LLC. All Rights
  *                         reserved.
  * Copyright (c) 2013      FUJITSU LIMITED.  All rights reserved.
- * Copyright (c) 2014-2016 Research Organization for Information Science
+ * Copyright (c) 2014-2017 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  * $COPYRIGHT$
  *
@@ -45,17 +45,16 @@ mca_coll_base_alltoallv_intra_basic_inplace(const void *rbuf, const int *rcounts
     int i, j, size, rank, err=MPI_SUCCESS;
     ompi_request_t *req;
     char *allocated_buffer, *tmp_buffer;
-    size_t max_size, rdtype_size;
-    OPAL_PTRDIFF_TYPE ext, gap;
+    size_t max_size;
+    OPAL_PTRDIFF_TYPE ext, gap = 0;
 
     /* Initialize. */
 
     size = ompi_comm_size(comm);
     rank = ompi_comm_rank(comm);
-    ompi_datatype_type_size(rdtype, &rdtype_size);
 
     /* If only one process, we're done. */
-    if (1 == size || 0 == rdtype_size) {
+    if (1 == size) {
         return MPI_SUCCESS;
     }
 
@@ -67,6 +66,10 @@ mca_coll_base_alltoallv_intra_basic_inplace(const void *rbuf, const int *rcounts
     }
     /* The gap will always be the same as we are working on the same datatype */
 
+    if (OPAL_UNLIKELY(0 == max_size)) {
+        return MPI_SUCCESS;
+    }
+
     /* Allocate a temporary buffer */
     allocated_buffer = calloc (max_size, 1);
     if (NULL == allocated_buffer) {
@@ -78,7 +81,7 @@ mca_coll_base_alltoallv_intra_basic_inplace(const void *rbuf, const int *rcounts
     /* in-place alltoallv slow algorithm (but works) */
     for (i = 0 ; i < size ; ++i) {
         for (j = i+1 ; j < size ; ++j) {
-            if (i == rank && rcounts[j]) {
+            if (i == rank && 0 != rcounts[j]) {
                 /* Copy the data into the temporary buffer */
                 err = ompi_datatype_copy_content_same_ddt (rdtype, rcounts[j],
                                                            tmp_buffer, (char *) rbuf + rdisps[j] * ext);
@@ -93,7 +96,7 @@ mca_coll_base_alltoallv_intra_basic_inplace(const void *rbuf, const int *rcounts
                                           j, MCA_COLL_BASE_TAG_ALLTOALLV, MCA_PML_BASE_SEND_STANDARD,
                                           comm));
                 if (MPI_SUCCESS != err) { goto error_hndl; }
-            } else if (j == rank && rcounts[i]) {
+            } else if (j == rank && 0 != rcounts[i]) {
                 /* Copy the data into the temporary buffer */
                 err = ompi_datatype_copy_content_same_ddt (rdtype, rcounts[i],
                                                            tmp_buffer, (char *) rbuf + rdisps[i] * ext);
diff --git a/ompi/mca/coll/libnbc/Makefile.am b/ompi/mca/coll/libnbc/Makefile.am
index 4d3e90186a9..83984b1185b 100644
--- a/ompi/mca/coll/libnbc/Makefile.am
+++ b/ompi/mca/coll/libnbc/Makefile.am
@@ -12,6 +12,8 @@
 # Copyright (c) 2010      Cisco Systems, Inc.  All rights reserved.
 # Copyright (c) 2013      Los Alamos National Security, LLC. All rights
 #                         reserved.
+# Copyright (c) 2017      Research Organization for Information Science
+#                         and Technology (RIST). All rights reserved.
 # $COPYRIGHT$
 #
 # Additional copyrights may follow
@@ -37,7 +39,6 @@ sources = \
 	nbc_ialltoallw.c \
 	nbc_ibarrier.c \
 	nbc_ibcast.c \
-	nbc_ibcast_inter.c \
 	nbc_iexscan.c \
 	nbc_igather.c \
 	nbc_igatherv.c \
diff --git a/ompi/mca/coll/libnbc/nbc.c b/ompi/mca/coll/libnbc/nbc.c
index 7949fe1b90f..da1f4548a04 100644
--- a/ompi/mca/coll/libnbc/nbc.c
+++ b/ompi/mca/coll/libnbc/nbc.c
@@ -10,7 +10,7 @@
  *                         rights reserved.
  * Copyright (c) 2015      Los Alamos National Security, LLC.  All rights
  *                         reserved.
- * Copyright (c) 2015-2016 Research Organization for Information Science
+ * Copyright (c) 2015-2017 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  *
  * Author(s): Torsten Hoefler <htor@cs.indiana.edu>
@@ -709,6 +709,25 @@ int NBC_Start(NBC_Handle *handle, NBC_Schedule *schedule) {
   return OMPI_SUCCESS;
 }
 
+int NBC_Schedule_request(NBC_Schedule *schedule, ompi_communicator_t *comm, ompi_coll_libnbc_module_t *module, ompi_request_t **request, void *tmpbuf) {
+  int res;
+  NBC_Handle *handle;
+  res = NBC_Init_handle (comm, &handle, module);
+  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
+    return res;
+  }
+  handle->tmpbuf = tmpbuf;
+
+  res = NBC_Start (handle, schedule);
+  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
+    NBC_Return_handle (handle);
+    return res;
+  }
+
+  *request = (ompi_request_t *) handle;
+  return OMPI_SUCCESS;
+}
+
 #ifdef NBC_CACHE_SCHEDULE
 void NBC_SchedCache_args_delete_key_dummy(void *k) {
     /* do nothing because the key and the data element are identical :-)
diff --git a/ompi/mca/coll/libnbc/nbc_iallgather.c b/ompi/mca/coll/libnbc/nbc_iallgather.c
index 9f0fea3706d..a1b512c1822 100644
--- a/ompi/mca/coll/libnbc/nbc_iallgather.c
+++ b/ompi/mca/coll/libnbc/nbc_iallgather.c
@@ -50,7 +50,6 @@ int ompi_coll_libnbc_iallgather(const void* sendbuf, int sendcount, MPI_Datatype
 #ifdef NBC_CACHE_SCHEDULE
   NBC_Allgather_args *args, *found, search;
 #endif
-  NBC_Handle *handle;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
 
   NBC_IN_PLACE(sendbuf, recvbuf, inplace);
@@ -147,20 +146,12 @@ int ompi_coll_libnbc_iallgather(const void* sendbuf, int sendcount, MPI_Datatype
   }
 #endif
 
-  res = NBC_Init_handle (comm, &handle, libnbc_module);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
     OBJ_RELEASE(schedule);
     return res;
   }
 
-  res = NBC_Start (handle, schedule);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    OMPI_COLL_LIBNBC_REQUEST_RETURN(handle);
-    return res;
-  }
-
-  *request = (ompi_request_t *) handle;
-
   return OMPI_SUCCESS;
 }
 
@@ -172,7 +163,6 @@ int ompi_coll_libnbc_iallgather_inter(const void* sendbuf, int sendcount, MPI_Da
   MPI_Aint rcvext;
   NBC_Schedule *schedule;
   char *rbuf;
-  NBC_Handle *handle;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
 
   res = ompi_datatype_type_extent(recvtype, &rcvext);
@@ -213,19 +203,11 @@ int ompi_coll_libnbc_iallgather_inter(const void* sendbuf, int sendcount, MPI_Da
     return res;
   }
 
-  res = NBC_Init_handle (comm, &handle, libnbc_module);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
     OBJ_RELEASE(schedule);
     return res;
   }
 
-  res = NBC_Start (handle, schedule);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    OMPI_COLL_LIBNBC_REQUEST_RETURN(handle);
-    return res;
-  }
-
-  *request = (ompi_request_t *) handle;
-
   return OMPI_SUCCESS;
 }
diff --git a/ompi/mca/coll/libnbc/nbc_iallgatherv.c b/ompi/mca/coll/libnbc/nbc_iallgatherv.c
index ea3c3634403..4b9697f9040 100644
--- a/ompi/mca/coll/libnbc/nbc_iallgatherv.c
+++ b/ompi/mca/coll/libnbc/nbc_iallgatherv.c
@@ -11,7 +11,7 @@
  * Copyright (c) 2012      Oracle and/or its affiliates.  All rights reserved.
  * Copyright (c) 2013-2015 Los Alamos National Security, LLC. All rights
  *                         reserved.
- * Copyright (c) 2014-2016 Research Organization for Information Science
+ * Copyright (c) 2014-2017 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  *
  */
@@ -37,7 +37,6 @@ int ompi_coll_libnbc_iallgatherv(const void* sendbuf, int sendcount, MPI_Datatyp
   MPI_Aint rcvext;
   NBC_Schedule *schedule;
   char *rbuf, *sbuf, inplace;
-  NBC_Handle *handle;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
 
   NBC_IN_PLACE(sendbuf, recvbuf, inplace);
@@ -96,20 +95,12 @@ int ompi_coll_libnbc_iallgatherv(const void* sendbuf, int sendcount, MPI_Datatyp
     return res;
   }
 
-  res = NBC_Init_handle (comm, &handle, libnbc_module);
+  res = NBC_Schedule_request (schedule, comm, libnbc_module, request, NULL);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
     OBJ_RELEASE(schedule);
     return res;
   }
 
-  res = NBC_Start (handle, schedule);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
-    return res;
-  }
-
-  *request = (ompi_request_t *) handle;
-
   return OMPI_SUCCESS;
 }
 
@@ -120,7 +111,6 @@ int ompi_coll_libnbc_iallgatherv_inter(const void* sendbuf, int sendcount, MPI_D
   int res, rsize;
   MPI_Aint rcvext;
   NBC_Schedule *schedule;
-  NBC_Handle *handle;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
 
   rsize = ompi_comm_remote_size (comm);
@@ -165,19 +155,11 @@ int ompi_coll_libnbc_iallgatherv_inter(const void* sendbuf, int sendcount, MPI_D
     return res;
   }
 
-  res = NBC_Init_handle (comm, &handle, libnbc_module);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
     OBJ_RELEASE(schedule);
     return res;
   }
 
-  res = NBC_Start (handle, schedule);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
-    return res;
-  }
-
-  *request = (ompi_request_t *) handle;
-
   return OMPI_SUCCESS;
 }
diff --git a/ompi/mca/coll/libnbc/nbc_iallreduce.c b/ompi/mca/coll/libnbc/nbc_iallreduce.c
index 0b624f90c44..f444c6aa716 100644
--- a/ompi/mca/coll/libnbc/nbc_iallreduce.c
+++ b/ompi/mca/coll/libnbc/nbc_iallreduce.c
@@ -21,13 +21,13 @@
 #include <assert.h>
 
 static inline int allred_sched_diss(int rank, int p, int count, MPI_Datatype datatype, ptrdiff_t gap, const void *sendbuf,
-                                    void *recvbuf, MPI_Op op, char inplace, NBC_Schedule *schedule, NBC_Handle *handle);
+                                    void *recvbuf, MPI_Op op, char inplace, NBC_Schedule *schedule, void *tmpbuf);
 static inline int allred_sched_ring(int rank, int p, int count, MPI_Datatype datatype, const void *sendbuf,
                                     void *recvbuf, MPI_Op op, int size, int ext, NBC_Schedule *schedule,
-                                    NBC_Handle *handle);
+                                    void *tmpbuf);
 static inline int allred_sched_linear(int rank, int p, const void *sendbuf, void *recvbuf, int count,
                                       MPI_Datatype datatype, ptrdiff_t gap, MPI_Op op, int ext, int size,
-                                      NBC_Schedule *schedule, NBC_Handle *handle);
+                                      NBC_Schedule *schedule, void *tmpbuf);
 
 #ifdef NBC_CACHE_SCHEDULE
 /* tree comparison function for schedule cache */
@@ -61,7 +61,7 @@ int ompi_coll_libnbc_iallreduce(const void* sendbuf, void* recvbuf, int count, M
 #endif
   enum { NBC_ARED_BINOMIAL, NBC_ARED_RING } alg;
   char inplace;
-  NBC_Handle *handle;
+  void *tmpbuf = NULL;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
   ptrdiff_t span, gap;
 
@@ -82,27 +82,24 @@ int ompi_coll_libnbc_iallreduce(const void* sendbuf, void* recvbuf, int count, M
     return res;
   }
 
-  res = NBC_Init_handle (comm, &handle, libnbc_module);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    return res;
+  if (1 == p) {
+    if (!inplace) {
+      /* for a single node - copy data to receivebuf */
+      res = NBC_Copy(sendbuf, count, datatype, recvbuf, count, datatype, comm);
+      if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
+        return res;
+      }
+    }
+    *request = &ompi_request_empty;
+    return OMPI_SUCCESS;
   }
 
   span = opal_datatype_span(&datatype->super, count, &gap);
-  handle->tmpbuf = malloc (span);
-  if (OPAL_UNLIKELY(NULL == handle->tmpbuf)) {
-    NBC_Return_handle (handle);
+  tmpbuf = malloc (span);
+  if (OPAL_UNLIKELY(NULL == tmpbuf)) {
     return OMPI_ERR_OUT_OF_RESOURCE;
   }
 
-  if ((p == 1) && !inplace) {
-    /* for a single node - copy data to receivebuf */
-    res = NBC_Copy(sendbuf, count, datatype, recvbuf, count, datatype, comm);
-    if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-      NBC_Return_handle (handle);
-      return res;
-    }
-  }
-
   /* algorithm selection */
   if(p < 4 || size*count < 65536 || !ompi_op_is_commute(op) || inplace) {
     alg = NBC_ARED_BINOMIAL;
@@ -122,30 +119,29 @@ int ompi_coll_libnbc_iallreduce(const void* sendbuf, void* recvbuf, int count, M
 #endif
     schedule = OBJ_NEW(NBC_Schedule);
     if (NULL == schedule) {
-      NBC_Return_handle (handle);
+      free(tmpbuf);
       return OMPI_ERR_OUT_OF_RESOURCE;
     }
 
-    /* ensure the schedule is released with the handle on error */
-    handle->schedule = schedule;
-
     switch(alg) {
       case NBC_ARED_BINOMIAL:
-        res = allred_sched_diss(rank, p, count, datatype, gap, sendbuf, recvbuf, op, inplace, schedule, handle);
+        res = allred_sched_diss(rank, p, count, datatype, gap, sendbuf, recvbuf, op, inplace, schedule, tmpbuf);
         break;
       case NBC_ARED_RING:
-        res = allred_sched_ring(rank, p, count, datatype, sendbuf, recvbuf, op, size, ext, schedule, handle);
+        res = allred_sched_ring(rank, p, count, datatype, sendbuf, recvbuf, op, size, ext, schedule, tmpbuf);
         break;
     }
 
     if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-      NBC_Return_handle (handle);
+      OBJ_RELEASE(schedule);
+      free(tmpbuf);
       return res;
     }
 
     res = NBC_Sched_commit(schedule);
     if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-      NBC_Return_handle (handle);
+      OBJ_RELEASE(schedule);
+      free(tmpbuf);
       return res;
     }
 
@@ -180,15 +176,13 @@ int ompi_coll_libnbc_iallreduce(const void* sendbuf, void* recvbuf, int count, M
   }
 #endif
 
-  res = NBC_Start (handle, schedule);
+  res = NBC_Schedule_request (schedule, comm, libnbc_module, request, tmpbuf);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
+    OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return res;
   }
 
-  *request = (ompi_request_t *) handle;
-
-  /* tmpbuf is freed with the handle */
   return OMPI_SUCCESS;
 }
 
@@ -200,7 +194,7 @@ int ompi_coll_libnbc_iallreduce_inter(const void* sendbuf, void* recvbuf, int co
   size_t size;
   MPI_Aint ext;
   NBC_Schedule *schedule;
-  NBC_Handle *handle;
+  void *tmpbuf = NULL;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
   ptrdiff_t span, gap;
 
@@ -219,49 +213,40 @@ int ompi_coll_libnbc_iallreduce_inter(const void* sendbuf, void* recvbuf, int co
     return res;
   }
 
-  res = NBC_Init_handle (comm, &handle, libnbc_module);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    return res;
-  }
-
   span = opal_datatype_span(&datatype->super, count, &gap);
-  handle->tmpbuf = malloc (span);
-  if (OPAL_UNLIKELY(NULL == handle->tmpbuf)) {
-    NBC_Return_handle (handle);
+  tmpbuf = malloc (span);
+  if (OPAL_UNLIKELY(NULL == tmpbuf)) {
     return OMPI_ERR_OUT_OF_RESOURCE;
   }
 
   schedule = OBJ_NEW(NBC_Schedule);
   if (OPAL_UNLIKELY(NULL == schedule)) {
-    NBC_Return_handle (handle);
+    free(tmpbuf);
     return OMPI_ERR_OUT_OF_RESOURCE;
   }
 
-  /* ensure the schedule is released with the handle on error */
-  handle->schedule = schedule;
-
   res = allred_sched_linear (rank, rsize, sendbuf, recvbuf, count, datatype, gap, op,
-                             ext, size, schedule, handle);
+                             ext, size, schedule, tmpbuf);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
+    OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return res;
   }
 
   res = NBC_Sched_commit(schedule);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
+    OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return res;
   }
 
-  res = NBC_Start(handle, schedule);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, tmpbuf);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
+    OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return res;
   }
 
-  *request = (ompi_request_t *) handle;
-
-  /* tmpbuf is freed with the handle */
   return OMPI_SUCCESS;
 }
 
@@ -302,7 +287,7 @@ int ompi_coll_libnbc_iallreduce_inter(const void* sendbuf, void* recvbuf, int co
   if (vrank == root) rank = 0; \
 }
 static inline int allred_sched_diss(int rank, int p, int count, MPI_Datatype datatype, ptrdiff_t gap, const void *sendbuf, void *recvbuf,
-                                    MPI_Op op, char inplace, NBC_Schedule *schedule, NBC_Handle *handle) {
+                                    MPI_Op op, char inplace, NBC_Schedule *schedule, void *tmpbuf) {
   int root, vrank, maxr, vpeer, peer, res;
   char *rbuf, *lbuf, *buf;
   int tmprbuf, tmplbuf;
@@ -322,7 +307,7 @@ static inline int allred_sched_diss(int rank, int p, int count, MPI_Datatype dat
     rbuf = recvbuf;
     tmprbuf = false;
     if (inplace) {
-        res = NBC_Copy(rbuf, count, datatype, ((char *)handle->tmpbuf) - gap, count, datatype, MPI_COMM_SELF);
+        res = NBC_Copy(rbuf, count, datatype, ((char *)tmpbuf) - gap, count, datatype, MPI_COMM_SELF);
         if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
           return res;
         }
@@ -341,7 +326,7 @@ static inline int allred_sched_diss(int rank, int p, int count, MPI_Datatype dat
           return res;
         }
 
-        /* this cannot be done until handle->tmpbuf is unused :-( so barrier after the op */
+        /* this cannot be done until tmpbuf is unused :-( so barrier after the op */
         if (firstred && !inplace) {
           /* perform the reduce with the senbuf */
           res = NBC_Sched_op (sendbuf, false, rbuf, tmprbuf, count, datatype, op, schedule, true);
@@ -417,7 +402,7 @@ static inline int allred_sched_diss(int rank, int p, int count, MPI_Datatype dat
 }
 
 static inline int allred_sched_ring (int r, int p, int count, MPI_Datatype datatype, const void *sendbuf, void *recvbuf, MPI_Op op,
-                                     int size, int ext, NBC_Schedule *schedule, NBC_Handle *handle) {
+                                     int size, int ext, NBC_Schedule *schedule, void *tmpbuf) {
   int segsize, *segsizes, *segoffsets; /* segment sizes and offsets per segment (number of segments == number of nodes */
   int speer, rpeer; /* send and recvpeer */
   int res = OMPI_SUCCESS;
@@ -617,7 +602,7 @@ static inline int allred_sched_ring (int r, int p, int count, MPI_Datatype datat
 }
 
 static inline int allred_sched_linear(int rank, int rsize, const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype,
-				      ptrdiff_t gap, MPI_Op op, int ext, int size, NBC_Schedule *schedule, NBC_Handle *handle) {
+				      ptrdiff_t gap, MPI_Op op, int ext, int size, NBC_Schedule *schedule, void *tmpbuf) {
   int res;
 
   if (0 == count) {
diff --git a/ompi/mca/coll/libnbc/nbc_ialltoall.c b/ompi/mca/coll/libnbc/nbc_ialltoall.c
index c9df894cc01..4cf83ae6306 100644
--- a/ompi/mca/coll/libnbc/nbc_ialltoall.c
+++ b/ompi/mca/coll/libnbc/nbc_ialltoall.c
@@ -8,7 +8,7 @@
  * Copyright (c) 2013-2015 Los Alamos National Security, LLC. All rights
  *                         reserved.
  * Copyright (c) 2014      NVIDIA Corporation.  All rights reserved.
- * Copyright (c) 2014-2016 Research Organization for Information Science
+ * Copyright (c) 2014-2017 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  *
  * Author(s): Torsten Hoefler <htor@cs.indiana.edu>
@@ -24,7 +24,7 @@ static inline int a2a_sched_pairwise(int rank, int p, MPI_Aint sndext, MPI_Aint
                                      int recvcount, MPI_Datatype recvtype, MPI_Comm comm);
 static inline int a2a_sched_diss(int rank, int p, MPI_Aint sndext, MPI_Aint rcvext, NBC_Schedule* schedule,
                                  const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf,
-                                 int recvcount, MPI_Datatype recvtype, MPI_Comm comm, NBC_Handle *handle);
+                                 int recvcount, MPI_Datatype recvtype, MPI_Comm comm, void* tmpbuf);
 static inline int a2a_sched_inplace(int rank, int p, NBC_Schedule* schedule, void* buf, int count,
                                    MPI_Datatype type, MPI_Aint ext, ptrdiff_t gap, MPI_Comm comm);
 
@@ -62,7 +62,7 @@ int ompi_coll_libnbc_ialltoall(const void* sendbuf, int sendcount, MPI_Datatype
 #endif
   char *rbuf, *sbuf, inplace;
   enum {NBC_A2A_LINEAR, NBC_A2A_PAIRWISE, NBC_A2A_DISS, NBC_A2A_INPLACE} alg;
-  NBC_Handle *handle;
+  void *tmpbuf = NULL;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
   ptrdiff_t span, gap;
 
@@ -115,17 +115,11 @@ int ompi_coll_libnbc_ialltoall(const void* sendbuf, int sendcount, MPI_Datatype
     }
   }
 
-  res = NBC_Init_handle(comm, &handle, libnbc_module);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    return res;
-  }
-
   /* allocate temp buffer if we need one */
   if (alg == NBC_A2A_INPLACE) {
     span = opal_datatype_span(&recvtype->super, recvcount, &gap);
-    handle->tmpbuf = malloc(span);
-    if (OPAL_UNLIKELY(NULL == handle->tmpbuf)) {
-      NBC_Return_handle (handle);
+    tmpbuf = malloc(span);
+    if (OPAL_UNLIKELY(NULL == tmpbuf)) {
       return OMPI_ERR_OUT_OF_RESOURCE;
     }
   } else if (alg == NBC_A2A_DISS) {
@@ -136,21 +130,19 @@ int ompi_coll_libnbc_ialltoall(const void* sendbuf, int sendcount, MPI_Datatype
       res = PMPI_Pack_size (sendcount, sendtype, comm, &datasize);
       if (MPI_SUCCESS != res) {
         NBC_Error("MPI Error in PMPI_Pack_size() (%i)", res);
-        NBC_Return_handle (handle);
         return res;
       }
     }
 
     /* allocate temporary buffers */
     if ((p & 1) == 0) {
-      handle->tmpbuf = malloc (datasize * p * 2);
+      tmpbuf = malloc (datasize * p * 2);
     } else {
       /* we cannot divide p by two, so alloc more to be safe ... */
-      handle->tmpbuf = malloc (datasize * (p / 2 + 1) * 2 * 2);
+      tmpbuf = malloc (datasize * (p / 2 + 1) * 2 * 2);
     }
 
-    if (OPAL_UNLIKELY(NULL == handle->tmpbuf)) {
-      NBC_Return_handle (handle);
+    if (OPAL_UNLIKELY(NULL == tmpbuf)) {
       return OMPI_ERR_OUT_OF_RESOURCE;
     }
 
@@ -161,29 +153,29 @@ int ompi_coll_libnbc_ialltoall(const void* sendbuf, int sendcount, MPI_Datatype
     if (NBC_Type_intrinsic(sendtype)) {
 #endif /* OPAL_CUDA_SUPPORT */
       /* contiguous - just copy (1st copy) */
-      memcpy (handle->tmpbuf, (char *) sendbuf + datasize * rank, datasize * (p - rank));
+      memcpy (tmpbuf, (char *) sendbuf + datasize * rank, datasize * (p - rank));
       if (rank != 0) {
-        memcpy ((char *) handle->tmpbuf + datasize * (p - rank), sendbuf, datasize * rank);
+        memcpy ((char *) tmpbuf + datasize * (p - rank), sendbuf, datasize * rank);
       }
     } else {
       int pos=0;
 
       /* non-contiguous - pack */
-      res = PMPI_Pack ((char *) sendbuf + rank * sendcount * sndext, (p - rank) * sendcount, sendtype, handle->tmpbuf,
+      res = PMPI_Pack ((char *) sendbuf + rank * sendcount * sndext, (p - rank) * sendcount, sendtype, tmpbuf,
                       (p - rank) * datasize, &pos, comm);
       if (OPAL_UNLIKELY(MPI_SUCCESS != res)) {
         NBC_Error("MPI Error in PMPI_Pack() (%i)", res);
-        NBC_Return_handle (handle);
+        free(tmpbuf);
         return res;
       }
 
       if (rank != 0) {
         pos = 0;
-        res = PMPI_Pack(sendbuf, rank * sendcount, sendtype, (char *) handle->tmpbuf + datasize * (p - rank),
+        res = PMPI_Pack(sendbuf, rank * sendcount, sendtype, (char *) tmpbuf + datasize * (p - rank),
                        rank * datasize, &pos, comm);
         if (OPAL_UNLIKELY(MPI_SUCCESS != res)) {
           NBC_Error("MPI Error in PMPI_Pack() (%i)", res);
-          NBC_Return_handle (handle);
+          free(tmpbuf);
           return res;
         }
       }
@@ -204,13 +196,10 @@ int ompi_coll_libnbc_ialltoall(const void* sendbuf, int sendcount, MPI_Datatype
     /* not found - generate new schedule */
     schedule = OBJ_NEW(NBC_Schedule);
     if (OPAL_UNLIKELY(NULL == schedule)) {
-      NBC_Return_handle (handle);
+      free(tmpbuf);
       return OMPI_ERR_OUT_OF_RESOURCE;
     }
 
-    /* ensure the schedule is released with the handle on error */
-    handle->schedule = schedule;
-
     switch(alg) {
       case NBC_A2A_INPLACE:
         res = a2a_sched_inplace(rank, p, schedule, recvbuf, recvcount, recvtype, rcvext, gap, comm);
@@ -219,7 +208,7 @@ int ompi_coll_libnbc_ialltoall(const void* sendbuf, int sendcount, MPI_Datatype
         res = a2a_sched_linear(rank, p, sndext, rcvext, schedule, sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, comm);
         break;
       case NBC_A2A_DISS:
-        res = a2a_sched_diss(rank, p, sndext, rcvext, schedule, sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, comm, handle);
+        res = a2a_sched_diss(rank, p, sndext, rcvext, schedule, sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, comm, tmpbuf);
         break;
       case NBC_A2A_PAIRWISE:
         res = a2a_sched_pairwise(rank, p, sndext, rcvext, schedule, sendbuf, sendcount, sendtype, recvbuf, recvcount, recvtype, comm);
@@ -227,13 +216,15 @@ int ompi_coll_libnbc_ialltoall(const void* sendbuf, int sendcount, MPI_Datatype
     }
 
     if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-      NBC_Return_handle (handle);
+      OBJ_RELEASE(schedule);
+      free(tmpbuf);
       return res;
     }
 
     res = NBC_Sched_commit(schedule);
     if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-      NBC_Return_handle (handle);
+      OBJ_RELEASE(schedule);
+      free(tmpbuf);
       return res;
     }
 
@@ -269,14 +260,13 @@ int ompi_coll_libnbc_ialltoall(const void* sendbuf, int sendcount, MPI_Datatype
   }
 #endif
 
-  res = NBC_Start (handle, schedule);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, tmpbuf);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
+    OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return res;
   }
 
-  *request = (ompi_request_t *) handle;
-
   return OMPI_SUCCESS;
 }
 
@@ -288,7 +278,6 @@ int ompi_coll_libnbc_ialltoall_inter (const void* sendbuf, int sendcount, MPI_Da
   MPI_Aint sndext, rcvext;
   NBC_Schedule *schedule;
   char *rbuf, *sbuf;
-  NBC_Handle *handle;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
 
   rsize = ompi_comm_remote_size (comm);
@@ -337,20 +326,12 @@ int ompi_coll_libnbc_ialltoall_inter (const void* sendbuf, int sendcount, MPI_Da
     return res;
   }
 
-  res = NBC_Init_handle (comm, &handle, libnbc_module);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
     OBJ_RELEASE(schedule);
     return res;
   }
 
-  res = NBC_Start (handle, schedule);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
-    return res;
-  }
-
-  *request = (ompi_request_t *) handle;
-
   return OMPI_SUCCESS;
 }
 
@@ -413,7 +394,7 @@ static inline int a2a_sched_linear(int rank, int p, MPI_Aint sndext, MPI_Aint rc
 
 static inline int a2a_sched_diss(int rank, int p, MPI_Aint sndext, MPI_Aint rcvext, NBC_Schedule* schedule,
                                  const void* sendbuf, int sendcount, MPI_Datatype sendtype, void* recvbuf, int recvcount,
-                                 MPI_Datatype recvtype, MPI_Comm comm, NBC_Handle *handle) {
+                                 MPI_Datatype recvtype, MPI_Comm comm, void* tmpbuf) {
   int res, speer, rpeer, datasize, offset, virtp;
   char *rbuf, *rtmpbuf, *stmpbuf;
 
@@ -433,13 +414,13 @@ static inline int a2a_sched_diss(int rank, int p, MPI_Aint sndext, MPI_Aint rcve
 
   /* allocate temporary buffers */
   if ((p & 1) == 0) {
-    rtmpbuf = (char *) handle->tmpbuf + datasize * p;
-    stmpbuf = (char *) handle->tmpbuf + datasize * (p + p / 2);
+    rtmpbuf = (char *)tmpbuf + datasize * p;
+    stmpbuf = (char *)tmpbuf + datasize * (p + p / 2);
   } else {
     /* we cannot divide p by two, so alloc more to be safe ... */
     virtp = (p / 2 + 1) * 2;
-    rtmpbuf = (char *) handle->tmpbuf + datasize * p;
-    stmpbuf = (char *) handle->tmpbuf + datasize * (p + virtp / 2);
+    rtmpbuf = (char *)tmpbuf + datasize * p;
+    stmpbuf = (char *)tmpbuf + datasize * (p + virtp / 2);
   }
 
   /* phase 2 - communicate */
@@ -451,7 +432,7 @@ static inline int a2a_sched_diss(int rank, int p, MPI_Aint sndext, MPI_Aint rcve
         /* copy data to sendbuffer (2nd copy) - could be avoided using iovecs */
         /*printf("[%i] round %i: copying element %i to buffer %lu\n", rank, r, i, (unsigned long)(stmpbuf+offset));*/
         res = NBC_Sched_copy((void *)(intptr_t)(i * datasize), true, datasize, MPI_BYTE, stmpbuf + offset -
-                             (intptr_t) handle->tmpbuf, true, datasize, MPI_BYTE, schedule, false);
+                             (intptr_t)tmpbuf, true, datasize, MPI_BYTE, schedule, false);
         if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
           return res;
         }
@@ -463,12 +444,12 @@ static inline int a2a_sched_diss(int rank, int p, MPI_Aint sndext, MPI_Aint rcve
     /* add p because modulo does not work with negative values */
     rpeer = ((rank - r) + p) % p;
 
-    res = NBC_Sched_recv (rtmpbuf - (intptr_t) handle->tmpbuf, true, offset, MPI_BYTE, rpeer, schedule, false);
+    res = NBC_Sched_recv (rtmpbuf - (intptr_t)tmpbuf, true, offset, MPI_BYTE, rpeer, schedule, false);
     if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
       return res;
     }
 
-    res = NBC_Sched_send (stmpbuf - (intptr_t) handle->tmpbuf, true, offset, MPI_BYTE, speer, schedule, true);
+    res = NBC_Sched_send (stmpbuf - (intptr_t)tmpbuf, true, offset, MPI_BYTE, speer, schedule, true);
     if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
       return res;
     }
@@ -479,7 +460,7 @@ static inline int a2a_sched_diss(int rank, int p, MPI_Aint sndext, MPI_Aint rcve
       /* test if bit r is set in rank number i */
       if (i & r) {
         /* copy data to tmpbuffer (3rd copy) - could be avoided using iovecs */
-        res = NBC_Sched_copy (rtmpbuf + offset - (intptr_t) handle->tmpbuf, true, datasize, MPI_BYTE,
+        res = NBC_Sched_copy (rtmpbuf + offset - (intptr_t)tmpbuf, true, datasize, MPI_BYTE,
                               (void *)(intptr_t)(i * datasize), true, datasize, MPI_BYTE, schedule,
                               false);
         if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
@@ -491,8 +472,7 @@ static inline int a2a_sched_diss(int rank, int p, MPI_Aint sndext, MPI_Aint rcve
     }
   }
 
-  /* phase 3 - reorder - data is now in wrong order in handle->tmpbuf -
-   * reorder it into recvbuf */
+  /* phase 3 - reorder - data is now in wrong order in tmpbuf - reorder it into recvbuf */
   for (int i = 0 ; i < p; ++i) {
     rbuf = (char *) recvbuf + ((rank - i + p) % p) * recvcount * rcvext;
     res = NBC_Sched_unpack ((void *)(intptr_t) (i * datasize), true, recvcount, recvtype, rbuf, false, schedule,
diff --git a/ompi/mca/coll/libnbc/nbc_ialltoallv.c b/ompi/mca/coll/libnbc/nbc_ialltoallv.c
index d39838124d8..6d187b51b97 100644
--- a/ompi/mca/coll/libnbc/nbc_ialltoallv.c
+++ b/ompi/mca/coll/libnbc/nbc_ialltoallv.c
@@ -5,7 +5,7 @@
  *                         Corporation.  All rights reserved.
  * Copyright (c) 2006      The Technical University of Chemnitz. All
  *                         rights reserved.
- * Copyright (c) 2014-2016 Research Organization for Information Science
+ * Copyright (c) 2014-2017 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  * Copyright (c) 2015      Los Alamos National Security, LLC. All rights
  *                         reserved.
@@ -46,7 +46,7 @@ int ompi_coll_libnbc_ialltoallv(const void* sendbuf, const int *sendcounts, cons
   NBC_Schedule *schedule;
   char *rbuf, *sbuf, inplace;
   ptrdiff_t gap, span;
-  NBC_Handle *handle;
+  void * tmpbuf = NULL;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
 
   NBC_IN_PLACE(sendbuf, recvbuf, inplace);
@@ -60,11 +60,6 @@ int ompi_coll_libnbc_ialltoallv(const void* sendbuf, const int *sendcounts, cons
     return res;
   }
 
-  res = NBC_Init_handle (comm, &handle, libnbc_module);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    return res;
-  }
-
   /* copy data to receivbuffer */
   if (inplace) {
     int count = 0;
@@ -74,9 +69,12 @@ int ompi_coll_libnbc_ialltoallv(const void* sendbuf, const int *sendcounts, cons
       }
     }
     span = opal_datatype_span(&recvtype->super, count, &gap);
-    handle->tmpbuf = malloc(span);
-    if (OPAL_UNLIKELY(NULL == handle->tmpbuf)) {
-      NBC_Return_handle (handle);
+    if (OPAL_UNLIKELY(0 == span)) {
+      *request = &ompi_request_empty;
+      return MPI_SUCCESS;
+    }
+    tmpbuf = malloc(span);
+    if (OPAL_UNLIKELY(NULL == tmpbuf)) {
       return OMPI_ERR_OUT_OF_RESOURCE;
     }
     sendcounts = recvcounts;
@@ -99,7 +97,7 @@ int ompi_coll_libnbc_ialltoallv(const void* sendbuf, const int *sendcounts, cons
 
   schedule = OBJ_NEW(NBC_Schedule);
   if (OPAL_UNLIKELY(NULL == schedule)) {
-    NBC_Return_handle (handle);
+    free(tmpbuf);
     return OMPI_ERR_OUT_OF_RESOURCE;
   }
 
@@ -113,27 +111,25 @@ int ompi_coll_libnbc_ialltoallv(const void* sendbuf, const int *sendcounts, cons
                             recvbuf, recvcounts, rdispls, rcvext, recvtype);
   }
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
     OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return res;
   }
 
   res = NBC_Sched_commit (schedule);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
     OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return res;
   }
 
-  res = NBC_Start(handle, schedule);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, tmpbuf);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
     OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return res;
   }
 
-  *request = (ompi_request_t *) handle;
-
   return OMPI_SUCCESS;
 }
 
@@ -146,7 +142,6 @@ int ompi_coll_libnbc_ialltoallv_inter (const void* sendbuf, const int *sendcount
   int res, rsize;
   MPI_Aint sndext, rcvext;
   NBC_Schedule *schedule;
-  NBC_Handle *handle;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
 
 
@@ -196,21 +191,12 @@ int ompi_coll_libnbc_ialltoallv_inter (const void* sendbuf, const int *sendcount
     return res;
   }
 
-  res = NBC_Init_handle(comm, &handle, libnbc_module);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
     OBJ_RELEASE(schedule);
     return res;
   }
 
-  res = NBC_Start(handle, schedule);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
-    OBJ_RELEASE(schedule);
-    return res;
-  }
-
-  *request = (ompi_request_t *) handle;
-
   return OMPI_SUCCESS;
 }
 
@@ -336,13 +322,15 @@ static inline int a2av_sched_inplace(int rank, int p, NBC_Schedule *schedule,
     if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
       return res;
     }
-    res = NBC_Sched_send ((void *)(-gap), true , counts[peer], type, peer, schedule, false);
-    if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-      return res;
-    }
-    res = NBC_Sched_recv (tbuf, false , counts[peer], type, peer, schedule, true);
-    if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-      return res;
+    if (0 != counts[peer]) {
+      res = NBC_Sched_send ((void *)(-gap), true , counts[peer], type, peer, schedule, false);
+      if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
+        return res;
+      }
+      res = NBC_Sched_recv (tbuf, false , counts[peer], type, peer, schedule, true);
+      if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
+        return res;
+      }
     }
   }
 
diff --git a/ompi/mca/coll/libnbc/nbc_ialltoallw.c b/ompi/mca/coll/libnbc/nbc_ialltoallw.c
index ec29a3a355f..1e095a65234 100644
--- a/ompi/mca/coll/libnbc/nbc_ialltoallw.c
+++ b/ompi/mca/coll/libnbc/nbc_ialltoallw.c
@@ -5,7 +5,7 @@
  *                         Corporation.  All rights reserved.
  * Copyright (c) 2006      The Technical University of Chemnitz. All
  *                         rights reserved.
- * Copyright (c) 2014-2016 Research Organization for Information Science
+ * Copyright (c) 2014-2017 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  * Copyright (c) 2015      Los Alamos National Security, LLC. All rights
  *                         reserved.
@@ -45,7 +45,7 @@ int ompi_coll_libnbc_ialltoallw(const void* sendbuf, const int *sendcounts, cons
   NBC_Schedule *schedule;
   char *rbuf, *sbuf, inplace;
   ptrdiff_t span=0;
-  NBC_Handle *handle;
+  void *tmpbuf = NULL;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
 
   NBC_IN_PLACE(sendbuf, recvbuf, inplace);
@@ -53,11 +53,6 @@ int ompi_coll_libnbc_ialltoallw(const void* sendbuf, const int *sendcounts, cons
   rank = ompi_comm_rank (comm);
   p = ompi_comm_size (comm);
 
-  res = NBC_Init_handle (comm, &handle, libnbc_module);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    return res;
-  }
-
   /* copy data to receivbuffer */
   if (inplace) {
     ptrdiff_t lgap, lspan;
@@ -67,9 +62,12 @@ int ompi_coll_libnbc_ialltoallw(const void* sendbuf, const int *sendcounts, cons
         span = lspan;
       }
     }
-    handle->tmpbuf = malloc(span);
-    if (OPAL_UNLIKELY(NULL == handle->tmpbuf)) {
-      NBC_Return_handle (handle);
+    if (OPAL_UNLIKELY(0 == span)) {
+      *request = &ompi_request_empty;
+      return OMPI_SUCCESS;
+    }
+    tmpbuf = malloc(span);
+    if (OPAL_UNLIKELY(NULL == tmpbuf)) {
       return OMPI_ERR_OUT_OF_RESOURCE;
     }
     sendcounts = recvcounts;
@@ -86,7 +84,7 @@ int ompi_coll_libnbc_ialltoallw(const void* sendbuf, const int *sendcounts, cons
 
   schedule = OBJ_NEW(NBC_Schedule);
   if (OPAL_UNLIKELY(NULL == schedule)) {
-    NBC_Return_handle (handle);
+    free(tmpbuf);
     return OMPI_ERR_OUT_OF_RESOURCE;
   }
 
@@ -99,26 +97,25 @@ int ompi_coll_libnbc_ialltoallw(const void* sendbuf, const int *sendcounts, cons
                             recvbuf, recvcounts, rdispls, recvtypes);
   }
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
     OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return res;
   }
 
   res = NBC_Sched_commit (schedule);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
     OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return res;
   }
 
-  res = NBC_Start (handle, schedule);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, tmpbuf);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
+    OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return res;
   }
 
-  *request = (ompi_request_t *) handle;
-
   return OMPI_SUCCESS;
 } 
 
@@ -131,7 +128,6 @@ int ompi_coll_libnbc_ialltoallw_inter (const void* sendbuf, const int *sendcount
   int res, rsize;
   NBC_Schedule *schedule;
   char *rbuf, *sbuf;
-  NBC_Handle *handle;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
 
   rsize = ompi_comm_remote_size (comm);
@@ -168,20 +164,12 @@ int ompi_coll_libnbc_ialltoallw_inter (const void* sendbuf, const int *sendcount
     return res;
   }
 
-  res = NBC_Init_handle (comm, &handle, libnbc_module);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
     OBJ_RELEASE(schedule);
     return res;
   }
 
-  res = NBC_Start (handle, schedule);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
-    return res;
-  }
-
-  *request = (ompi_request_t *) handle;
-
   return OMPI_SUCCESS;
 }
 
@@ -193,12 +181,14 @@ static inline int a2aw_sched_linear(int rank, int p, NBC_Schedule *schedule,
   int res;
 
   for (int i = 0; i < p; i++) {
+    ptrdiff_t gap, span;
     if (i == rank) {
       continue;
     }
 
     /* post send */
-    if (sendcounts[i] != 0) {
+    span = opal_datatype_span(&sendtypes[i]->super, sendcounts[i], &gap);
+    if (OPAL_LIKELY(0 < span)) {
       char *sbuf = (char *) sendbuf + sdispls[i];
       res = NBC_Sched_send (sbuf, false, sendcounts[i], sendtypes[i], i, schedule, false);
       if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
@@ -206,7 +196,8 @@ static inline int a2aw_sched_linear(int rank, int p, NBC_Schedule *schedule,
       }
     }
     /* post receive */
-    if (recvcounts[i] != 0) {
+    span = opal_datatype_span(&recvtypes[i]->super, recvcounts[i], &gap);
+    if (OPAL_LIKELY(0 < span)) {
       char *rbuf = (char *) recvbuf + rdispls[i];
       res = NBC_Sched_recv (rbuf, false, recvcounts[i], recvtypes[i], i, schedule, false);
       if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
diff --git a/ompi/mca/coll/libnbc/nbc_ibarrier.c b/ompi/mca/coll/libnbc/nbc_ibarrier.c
index 9d2c2f2c450..97838c549c5 100644
--- a/ompi/mca/coll/libnbc/nbc_ibarrier.c
+++ b/ompi/mca/coll/libnbc/nbc_ibarrier.c
@@ -7,7 +7,7 @@
  *                         rights reserved.
  * Copyright (c) 2013-2015 Los Alamos National Security, LLC. All rights
  *                         reserved.
- * Copyright (c) 2014      Research Organization for Information Science
+ * Copyright (c) 2014-2017 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  * Copyright (c) 2015      Mellanox Technologies. All rights reserved.
  *
diff --git a/ompi/mca/coll/libnbc/nbc_ibcast.c b/ompi/mca/coll/libnbc/nbc_ibcast.c
index 840e6cdce9b..229a783ab63 100644
--- a/ompi/mca/coll/libnbc/nbc_ibcast.c
+++ b/ompi/mca/coll/libnbc/nbc_ibcast.c
@@ -52,7 +52,6 @@ int ompi_coll_libnbc_ibcast(void *buffer, int count, MPI_Datatype datatype, int
   NBC_Bcast_args *args, *found, search;
 #endif
   enum { NBC_BCAST_LINEAR, NBC_BCAST_BINOMIAL, NBC_BCAST_CHAIN } alg;
-  NBC_Handle *handle;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
 
   rank = ompi_comm_rank (comm);
@@ -155,20 +154,12 @@ int ompi_coll_libnbc_ibcast(void *buffer, int count, MPI_Datatype datatype, int
   }
 #endif
 
-  res = NBC_Init_handle (comm, &handle, libnbc_module);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
     OBJ_RELEASE(schedule);
     return res;
   }
 
-  res = NBC_Start (handle, schedule);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
-    return res;
-  }
-
-  *request = (ompi_request_t *) handle;
-
   return OMPI_SUCCESS;
 }
 
@@ -323,3 +314,55 @@ static inline int bcast_sched_chain(int rank, int p, int root, NBC_Schedule *sch
 
   return OMPI_SUCCESS;
 }
+
+int ompi_coll_libnbc_ibcast_inter(void *buffer, int count, MPI_Datatype datatype, int root,
+                                  struct ompi_communicator_t *comm, ompi_request_t ** request,
+                                  struct mca_coll_base_module_2_1_0_t *module) {
+  int res;
+  NBC_Schedule *schedule;
+  ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
+
+  schedule = OBJ_NEW(NBC_Schedule);
+  if (OPAL_UNLIKELY(NULL == schedule)) {
+    return OMPI_ERR_OUT_OF_RESOURCE;
+  }
+
+  if (root != MPI_PROC_NULL) {
+    /* send to all others */
+    if (root == MPI_ROOT) {
+      int remsize;
+
+      remsize = ompi_comm_remote_size (comm);
+
+      for (int peer = 0 ; peer < remsize ; ++peer) {
+        /* send msg to peer */
+        res = NBC_Sched_send (buffer, false, count, datatype, peer, schedule, false);
+        if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
+          OBJ_RELEASE(schedule);
+          return res;
+        }
+      }
+    } else {
+      /* recv msg from root */
+      res = NBC_Sched_recv (buffer, false, count, datatype, root, schedule, false);
+      if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
+        OBJ_RELEASE(schedule);
+        return res;
+      }
+    }
+  }
+
+  res = NBC_Sched_commit (schedule);
+  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
+    OBJ_RELEASE(schedule);
+    return res;
+  }
+
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
+  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
+    OBJ_RELEASE(schedule);
+    return res;
+  }
+
+  return OMPI_SUCCESS;
+}
diff --git a/ompi/mca/coll/libnbc/nbc_ibcast_inter.c b/ompi/mca/coll/libnbc/nbc_ibcast_inter.c
deleted file mode 100644
index 8188ec65732..00000000000
--- a/ompi/mca/coll/libnbc/nbc_ibcast_inter.c
+++ /dev/null
@@ -1,77 +0,0 @@
-/* -*- Mode: C; c-basic-offset:2 ; indent-tabs-mode:nil -*- */
-/*
- * Copyright (c) 2006      The Trustees of Indiana University and Indiana
- *                         University Research and Technology
- *                         Corporation.  All rights reserved.
- * Copyright (c) 2006      The Technical University of Chemnitz. All
- *                         rights reserved.
- * Copyright (c) 2014-2015 Research Organization for Information Science
- *                         and Technology (RIST). All rights reserved.
- * Copyright (c) 2015      Los Alamos National Security, LLC. All rights
- *                         reserved.
- *
- * Author(s): Torsten Hoefler <htor@cs.indiana.edu>
- *
- */
-#include "nbc_internal.h"
-
-int ompi_coll_libnbc_ibcast_inter(void *buffer, int count, MPI_Datatype datatype, int root,
-                                  struct ompi_communicator_t *comm, ompi_request_t ** request,
-                                  struct mca_coll_base_module_2_1_0_t *module) {
-  int res;
-  NBC_Schedule *schedule;
-  NBC_Handle *handle;
-  ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
-
-  schedule = OBJ_NEW(NBC_Schedule);
-  if (OPAL_UNLIKELY(NULL == schedule)) {
-    return OMPI_ERR_OUT_OF_RESOURCE;
-  }
-
-  if (root != MPI_PROC_NULL) {
-    /* send to all others */
-    if (root == MPI_ROOT) {
-      int remsize;
-
-      remsize = ompi_comm_remote_size (comm);
-
-      for (int peer = 0 ; peer < remsize ; ++peer) {
-        /* send msg to peer */
-        res = NBC_Sched_send (buffer, false, count, datatype, peer, schedule, false);
-        if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-          OBJ_RELEASE(schedule);
-          return res;
-        }
-      }
-    } else {
-      /* recv msg from root */
-      res = NBC_Sched_recv (buffer, false, count, datatype, root, schedule, false);
-      if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-        OBJ_RELEASE(schedule);
-        return res;
-      }
-    }
-  }
-
-  res = NBC_Sched_commit (schedule);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    OBJ_RELEASE(schedule);
-    return res;
-  }
-
-  res = NBC_Init_handle (comm, &handle, libnbc_module);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    OBJ_RELEASE(schedule);
-    return res;
-  }
-
-  res = NBC_Start(handle, schedule);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
-    return res;
-  }
-
-  *request = (ompi_request_t *) handle;
-
-  return OMPI_SUCCESS;
-}
diff --git a/ompi/mca/coll/libnbc/nbc_iexscan.c b/ompi/mca/coll/libnbc/nbc_iexscan.c
index d43edebcc60..d1245caa636 100644
--- a/ompi/mca/coll/libnbc/nbc_iexscan.c
+++ b/ompi/mca/coll/libnbc/nbc_iexscan.c
@@ -7,7 +7,7 @@
  *                         rights reserved.
  * Copyright (c) 2013-2015 Los Alamos National Security, LLC. All rights
  *                         reserved.
- * Copyright (c) 2014-2015 Research Organization for Information Science
+ * Copyright (c) 2014-2017 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  *
  * Author(s): Torsten Hoefler <htor@cs.indiana.edu>
@@ -51,7 +51,7 @@ int ompi_coll_libnbc_iexscan(const void* sendbuf, void* recvbuf, int count, MPI_
     NBC_Scan_args *args, *found, search;
 #endif
     char inplace;
-    NBC_Handle *handle;
+    void *tmpbuf = NULL;
     ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
 
     NBC_IN_PLACE(sendbuf, recvbuf, inplace);
@@ -59,25 +59,19 @@ int ompi_coll_libnbc_iexscan(const void* sendbuf, void* recvbuf, int count, MPI_
     rank = ompi_comm_rank (comm);
     p = ompi_comm_size (comm);
 
-    res = NBC_Init_handle(comm, &handle, libnbc_module);
-    if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-        return res;
-    }
-
     span = opal_datatype_span(&datatype->super, count, &gap);
     if (0 < rank) {
-        handle->tmpbuf = malloc(span);
-        if (handle->tmpbuf == NULL) {
-            NBC_Return_handle (handle);
+        tmpbuf = malloc(span);
+        if (NULL == tmpbuf) {
             return OMPI_ERR_OUT_OF_RESOURCE;
         }
         if (inplace) {
-            res = NBC_Copy(recvbuf, count, datatype, (char *)handle->tmpbuf-gap, count, datatype, comm);
+            res = NBC_Copy(recvbuf, count, datatype, (char *)tmpbuf-gap, count, datatype, comm);
         } else {
-            res = NBC_Copy(sendbuf, count, datatype, (char *)handle->tmpbuf-gap, count, datatype, comm);
+            res = NBC_Copy(sendbuf, count, datatype, (char *)tmpbuf-gap, count, datatype, comm);
         }
         if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-            NBC_Return_handle (handle);
+            free(tmpbuf);
             return res;
         }
     }
@@ -94,18 +88,16 @@ int ompi_coll_libnbc_iexscan(const void* sendbuf, void* recvbuf, int count, MPI_
 #endif
         schedule = OBJ_NEW(NBC_Schedule);
         if (OPAL_UNLIKELY(NULL == schedule)) {
-            NBC_Return_handle (handle);
+            free(tmpbuf);
             return OMPI_ERR_OUT_OF_RESOURCE;
         }
 
-        /* make sure the schedule is released with the handle on error */
-        handle->schedule = schedule;
-
         if (rank != 0) {
             res = NBC_Sched_recv (recvbuf, false, count, datatype, rank-1, schedule, false);
 
             if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-                NBC_Return_handle (handle);
+                OBJ_RELEASE(schedule);
+                free(tmpbuf);
                 return res;
             }
 
@@ -113,7 +105,8 @@ int ompi_coll_libnbc_iexscan(const void* sendbuf, void* recvbuf, int count, MPI_
                 /* we have to wait until we have the data */
                 res = NBC_Sched_barrier(schedule);
                 if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-                    NBC_Return_handle (handle);
+                    OBJ_RELEASE(schedule);
+                    free(tmpbuf);
                     return res;
                 }
 
@@ -121,14 +114,16 @@ int ompi_coll_libnbc_iexscan(const void* sendbuf, void* recvbuf, int count, MPI_
                                      datatype, op, schedule, true);
 
                 if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-                    NBC_Return_handle (handle);
+                    OBJ_RELEASE(schedule);
+                    free(tmpbuf);
                     return res;
                 }
 
                 /* send reduced data onward */
                 res = NBC_Sched_send ((void *)(-gap), true, count, datatype, rank + 1, schedule, false);
                 if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-                    NBC_Return_handle (handle);
+                    OBJ_RELEASE(schedule);
+                    free(tmpbuf);
                     return res;
                 }
             }
@@ -139,14 +134,16 @@ int ompi_coll_libnbc_iexscan(const void* sendbuf, void* recvbuf, int count, MPI_
               res = NBC_Sched_send (sendbuf, false, count, datatype, 1, schedule, false);
             }
             if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-                NBC_Return_handle (handle);
+                OBJ_RELEASE(schedule);
+                free(tmpbuf);
                 return res;
             }
         }
 
         res = NBC_Sched_commit(schedule);
         if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-            NBC_Return_handle (handle);
+            OBJ_RELEASE(schedule);
+            free(tmpbuf);
             return res;
         }
 
@@ -181,14 +178,12 @@ int ompi_coll_libnbc_iexscan(const void* sendbuf, void* recvbuf, int count, MPI_
     }
 #endif
 
-    res = NBC_Start (handle, schedule);
+    res = NBC_Schedule_request(schedule, comm, libnbc_module, request, tmpbuf);
     if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-        NBC_Return_handle (handle);
+        OBJ_RELEASE(schedule);
+        free(tmpbuf);
         return res;
     }
 
-    *request = (ompi_request_t *) handle;
-
-    /* tmpbuf is freed with the handle */
     return OMPI_SUCCESS;
 }
diff --git a/ompi/mca/coll/libnbc/nbc_igather.c b/ompi/mca/coll/libnbc/nbc_igather.c
index edb334a7ea1..72b85a439e2 100644
--- a/ompi/mca/coll/libnbc/nbc_igather.c
+++ b/ompi/mca/coll/libnbc/nbc_igather.c
@@ -8,7 +8,7 @@
  * Copyright (c) 2013      The University of Tennessee and The University
  *                         of Tennessee Research Foundation.  All rights
  *                         reserved.
- * Copyright (c) 2014-2016 Research Organization for Information Science
+ * Copyright (c) 2014-2017 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  * Copyright (c) 2015      Los Alamos National Security, LLC. All rights
  *                         reserved.
@@ -47,7 +47,6 @@ int ompi_coll_libnbc_igather(const void* sendbuf, int sendcount, MPI_Datatype se
   MPI_Aint rcvext = 0;
   NBC_Schedule *schedule;
   char *rbuf, inplace = 0;
-  NBC_Handle *handle;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
 
   rank = ompi_comm_rank (comm);
@@ -157,20 +156,12 @@ int ompi_coll_libnbc_igather(const void* sendbuf, int sendcount, MPI_Datatype se
   }
 #endif
 
-  res = NBC_Init_handle (comm, &handle, libnbc_module);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
     OBJ_RELEASE(schedule);
     return res;
   }
 
-  res = NBC_Start (handle, schedule);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
-    return res;
-  }
-
-  *request = (ompi_request_t *) handle;
-
   return OMPI_SUCCESS;
 }
 
@@ -182,7 +173,6 @@ int ompi_coll_libnbc_igather_inter (const void* sendbuf, int sendcount, MPI_Data
     MPI_Aint rcvext = 0;
     NBC_Schedule *schedule;
     char *rbuf;
-    NBC_Handle *handle;
     ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
 
     rsize = ompi_comm_remote_size (comm);
@@ -226,19 +216,11 @@ int ompi_coll_libnbc_igather_inter (const void* sendbuf, int sendcount, MPI_Data
       return res;
     }
 
-    res = NBC_Init_handle (comm, &handle, libnbc_module);
+    res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
     if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
       OBJ_RELEASE(schedule);
       return res;
     }
 
-    res = NBC_Start (handle, schedule);
-    if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-      NBC_Return_handle (handle);
-      return res;
-    }
-
-    *request = (ompi_request_t *) handle;
-
     return OMPI_SUCCESS;
 }
diff --git a/ompi/mca/coll/libnbc/nbc_igatherv.c b/ompi/mca/coll/libnbc/nbc_igatherv.c
index 1b55002c4b2..15ff1c3c3ef 100644
--- a/ompi/mca/coll/libnbc/nbc_igatherv.c
+++ b/ompi/mca/coll/libnbc/nbc_igatherv.c
@@ -8,7 +8,7 @@
  * Copyright (c) 2013      The University of Tennessee and The University
  *                         of Tennessee Research Foundation.  All rights
  *                         reserved.
- * Copyright (c) 2014-2016 Research Organization for Information Science
+ * Copyright (c) 2014-2017 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  * Copyright (c) 2015      Los Alamos National Security, LLC.  All rights
  *                         reserved.
@@ -32,7 +32,6 @@ int ompi_coll_libnbc_igatherv(const void* sendbuf, int sendcount, MPI_Datatype s
   MPI_Aint rcvext = 0;
   NBC_Schedule *schedule;
   char *rbuf, inplace = 0;
-  NBC_Handle *handle;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
 
   rank = ompi_comm_rank (comm);
@@ -92,20 +91,12 @@ int ompi_coll_libnbc_igatherv(const void* sendbuf, int sendcount, MPI_Datatype s
     return res;
   }
 
-  res = NBC_Init_handle (comm, &handle, libnbc_module);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
     OBJ_RELEASE(schedule);
     return res;
   }
 
-  res = NBC_Start (handle, schedule);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
-    return res;
-  }
-
-  *request = (ompi_request_t *) handle;
-
   return OMPI_SUCCESS;
 }
 
@@ -117,7 +108,6 @@ int ompi_coll_libnbc_igatherv_inter (const void* sendbuf, int sendcount, MPI_Dat
   MPI_Aint rcvext;
   NBC_Schedule *schedule;
   char *rbuf;
-  NBC_Handle *handle;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
 
   rsize = ompi_comm_remote_size (comm);
@@ -161,19 +151,11 @@ int ompi_coll_libnbc_igatherv_inter (const void* sendbuf, int sendcount, MPI_Dat
     return res;
   }
 
-  res = NBC_Init_handle (comm, &handle, libnbc_module);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
     OBJ_RELEASE(schedule);
     return res;
   }
 
-  res = NBC_Start (handle, schedule);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
-    return res;
-  }
-
-  *request = (ompi_request_t *) handle;
-
   return OMPI_SUCCESS;
 }
diff --git a/ompi/mca/coll/libnbc/nbc_ineighbor_allgather.c b/ompi/mca/coll/libnbc/nbc_ineighbor_allgather.c
index e9ae7b388a9..5ccdbc4356f 100644
--- a/ompi/mca/coll/libnbc/nbc_ineighbor_allgather.c
+++ b/ompi/mca/coll/libnbc/nbc_ineighbor_allgather.c
@@ -5,7 +5,7 @@
  *                         Corporation.  All rights reserved.
  * Copyright (c) 2006      The Technical University of Chemnitz. All
  *                         rights reserved.
- * Copyright (c) 2014-2015 Research Organization for Information Science
+ * Copyright (c) 2014-2017 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  * Copyright (c) 2015      Los Alamos National Security, LLC.  All rights
  *                         reserved.
@@ -44,7 +44,6 @@ int ompi_coll_libnbc_ineighbor_allgather(const void *sbuf, int scount, MPI_Datat
                                          ompi_request_t ** request, struct mca_coll_base_module_2_1_0_t *module) {
   int res, indegree, outdegree, *srcs, *dsts;
   MPI_Aint rcvext;
-  NBC_Handle *handle;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
   NBC_Schedule *schedule;
 
@@ -149,20 +148,11 @@ int ompi_coll_libnbc_ineighbor_allgather(const void *sbuf, int scount, MPI_Datat
   }
 #endif
 
-  res = NBC_Init_handle(comm, &handle, libnbc_module);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
     OBJ_RELEASE(schedule);
     return res;
   }
 
-  res = NBC_Start(handle, schedule);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
-    OBJ_RELEASE(schedule);
-    return res;
-  }
-
-  *request = (ompi_request_t *) handle;
-
   return OMPI_SUCCESS;
 }
diff --git a/ompi/mca/coll/libnbc/nbc_ineighbor_allgatherv.c b/ompi/mca/coll/libnbc/nbc_ineighbor_allgatherv.c
index 50d85ee406a..ee18644f8a5 100644
--- a/ompi/mca/coll/libnbc/nbc_ineighbor_allgatherv.c
+++ b/ompi/mca/coll/libnbc/nbc_ineighbor_allgatherv.c
@@ -5,7 +5,7 @@
  *                         Corporation.  All rights reserved.
  * Copyright (c) 2006      The Technical University of Chemnitz. All
  *                         rights reserved.
- * Copyright (c) 2014-2015 Research Organization for Information Science
+ * Copyright (c) 2014-2017 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  * Copyright (c) 2015      Los Alamos National Security, LLC.  All rights
  *                         reserved.
@@ -45,7 +45,6 @@ int ompi_coll_libnbc_ineighbor_allgatherv(const void *sbuf, int scount, MPI_Data
 					  struct mca_coll_base_module_2_1_0_t *module) {
   int res, indegree, outdegree, *srcs, *dsts;
   MPI_Aint rcvext;
-  NBC_Handle *handle;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
   NBC_Schedule *schedule;
 
@@ -151,20 +150,11 @@ int ompi_coll_libnbc_ineighbor_allgatherv(const void *sbuf, int scount, MPI_Data
   }
 #endif
 
-  res = NBC_Init_handle(comm, &handle, libnbc_module);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
     OBJ_RELEASE(schedule);
     return res;
   }
 
-  res = NBC_Start (handle, schedule);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
-    OBJ_RELEASE(schedule);
-    return OMPI_ERR_OUT_OF_RESOURCE;
-  }
-
-  *request = (ompi_request_t *) handle;
-
   return OMPI_SUCCESS;
 }
diff --git a/ompi/mca/coll/libnbc/nbc_ineighbor_alltoall.c b/ompi/mca/coll/libnbc/nbc_ineighbor_alltoall.c
index 3d8b34a1918..51e3231be87 100644
--- a/ompi/mca/coll/libnbc/nbc_ineighbor_alltoall.c
+++ b/ompi/mca/coll/libnbc/nbc_ineighbor_alltoall.c
@@ -5,7 +5,7 @@
  *                         Corporation.  All rights reserved.
  * Copyright (c) 2006      The Technical University of Chemnitz. All
  *                         rights reserved.
- * Copyright (c) 2014-2015 Research Organization for Information Science
+ * Copyright (c) 2014-2017 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  * Copyright (c) 2015      Los Alamos National Security, LLC. All rights
  *                         reserved.
@@ -41,7 +41,6 @@ int ompi_coll_libnbc_ineighbor_alltoall(const void *sbuf, int scount, MPI_Dataty
                                         ompi_request_t ** request, struct mca_coll_base_module_2_1_0_t *module) {
   int res, indegree, outdegree, *srcs, *dsts;
   MPI_Aint sndext, rcvext;
-  NBC_Handle *handle;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
   NBC_Schedule *schedule;
 
@@ -153,19 +152,11 @@ int ompi_coll_libnbc_ineighbor_alltoall(const void *sbuf, int scount, MPI_Dataty
   }
 #endif
 
-  res = NBC_Init_handle (comm, &handle, libnbc_module);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
     OBJ_RELEASE(schedule);
     return res;
   }
 
-  res = NBC_Start (handle, schedule);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
-    return OMPI_ERR_OUT_OF_RESOURCE;
-  }
-
-  *request = (ompi_request_t *) handle;
-
   return OMPI_SUCCESS;
 }
diff --git a/ompi/mca/coll/libnbc/nbc_ineighbor_alltoallv.c b/ompi/mca/coll/libnbc/nbc_ineighbor_alltoallv.c
index 52983b1632b..2b30ffd9be0 100644
--- a/ompi/mca/coll/libnbc/nbc_ineighbor_alltoallv.c
+++ b/ompi/mca/coll/libnbc/nbc_ineighbor_alltoallv.c
@@ -5,7 +5,7 @@
  *                         Corporation.  All rights reserved.
  * Copyright (c) 2006      The Technical University of Chemnitz. All
  *                         rights reserved.
- * Copyright (c) 2014-2015 Research Organization for Information Science
+ * Copyright (c) 2014-2017 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  * Copyright (c) 2015      Los Alamos National Security, LLC.  All rights
  *                         reserved.
@@ -45,7 +45,6 @@ int ompi_coll_libnbc_ineighbor_alltoallv(const void *sbuf, const int *scounts, c
                                          struct mca_coll_base_module_2_1_0_t *module) {
   int res, indegree, outdegree, *srcs, *dsts;
   MPI_Aint sndext, rcvext;
-  NBC_Handle *handle;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
   NBC_Schedule *schedule;
 
@@ -158,19 +157,11 @@ int ompi_coll_libnbc_ineighbor_alltoallv(const void *sbuf, const int *scounts, c
   }
 #endif
 
-  res = NBC_Init_handle(comm, &handle, libnbc_module);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
     OBJ_RELEASE(schedule);
     return res;
   }
 
-  res = NBC_Start (handle, schedule);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
-    return res;
-  }
-
-  *request = (ompi_request_t *) handle;
-
   return OMPI_SUCCESS;
 }
diff --git a/ompi/mca/coll/libnbc/nbc_ineighbor_alltoallw.c b/ompi/mca/coll/libnbc/nbc_ineighbor_alltoallw.c
index 8bab6a00811..8907e7cd3a2 100644
--- a/ompi/mca/coll/libnbc/nbc_ineighbor_alltoallw.c
+++ b/ompi/mca/coll/libnbc/nbc_ineighbor_alltoallw.c
@@ -5,7 +5,7 @@
  *                         Corporation.  All rights reserved.
  * Copyright (c) 2006      The Technical University of Chemnitz. All
  *                         rights reserved.
- * Copyright (c) 2014-2015 Research Organization for Information Science
+ * Copyright (c) 2014-2017 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  * Copyright (c) 2015      Los Alamos National Security, LLC.  All rights
  *                         reserved.
@@ -43,7 +43,6 @@ int ompi_coll_libnbc_ineighbor_alltoallw(const void *sbuf, const int *scounts, c
                                          struct ompi_communicator_t *comm, ompi_request_t ** request,
                                          struct mca_coll_base_module_2_1_0_t *module) {
   int res, indegree, outdegree, *srcs, *dsts;
-  NBC_Handle *handle;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
   NBC_Schedule *schedule;
 
@@ -143,19 +142,11 @@ int ompi_coll_libnbc_ineighbor_alltoallw(const void *sbuf, const int *scounts, c
   }
 #endif
 
-  res = NBC_Init_handle(comm, &handle, libnbc_module);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
     OBJ_RELEASE(schedule);
     return res;
   }
 
-  res = NBC_Start(handle, schedule);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
-    return res;
-  }
-
-  *request = (ompi_request_t *) handle;
-
   return OMPI_SUCCESS;
 }
diff --git a/ompi/mca/coll/libnbc/nbc_internal.h b/ompi/mca/coll/libnbc/nbc_internal.h
index 915e4e232a9..33ad6dc4a4a 100644
--- a/ompi/mca/coll/libnbc/nbc_internal.h
+++ b/ompi/mca/coll/libnbc/nbc_internal.h
@@ -258,6 +258,7 @@ void NBC_SchedCache_args_delete_key_dummy(void *k);
 
 int NBC_Start(NBC_Handle *handle, NBC_Schedule *schedule);
 int NBC_Init_handle(struct ompi_communicator_t *comm, ompi_coll_libnbc_request_t **request, ompi_coll_libnbc_module_t *module);
+int NBC_Schedule_request(NBC_Schedule *schedule, ompi_communicator_t *comm, ompi_coll_libnbc_module_t *module, ompi_request_t **request, void *tmpbuf);
 void NBC_Return_handle(ompi_coll_libnbc_request_t *request);
 static inline int NBC_Type_intrinsic(MPI_Datatype type);
 int NBC_Create_fortran_handle(int *fhandle, NBC_Handle **handle);
diff --git a/ompi/mca/coll/libnbc/nbc_ireduce.c b/ompi/mca/coll/libnbc/nbc_ireduce.c
index cf78d073b16..c2e730921e8 100644
--- a/ompi/mca/coll/libnbc/nbc_ireduce.c
+++ b/ompi/mca/coll/libnbc/nbc_ireduce.c
@@ -20,12 +20,12 @@
 #include "nbc_internal.h"
 
 static inline int red_sched_binomial (int rank, int p, int root, const void *sendbuf, void *redbuf, char tmpredbuf, int count, MPI_Datatype datatype,
-                                      MPI_Op op, char inplace, NBC_Schedule *schedule, NBC_Handle *handle);
+                                      MPI_Op op, char inplace, NBC_Schedule *schedule, void *tmpbuf);
 static inline int red_sched_chain (int rank, int p, int root, const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype,
-                                   MPI_Op op, int ext, size_t size, NBC_Schedule *schedule, NBC_Handle *handle, int fragsize);
+                                   MPI_Op op, int ext, size_t size, NBC_Schedule *schedule, void *tmpbuf, int fragsize);
 
 static inline int red_sched_linear (int rank, int rsize, int root, const void *sendbuf, void *recvbuf, void *tmpbuf, int count, MPI_Datatype datatype,
-                                    MPI_Op op, NBC_Schedule *schedule, NBC_Handle *handle);
+                                    MPI_Op op, NBC_Schedule *schedule);
 
 #ifdef NBC_CACHE_SCHEDULE
 /* tree comparison function for schedule cache */
@@ -56,9 +56,9 @@ int ompi_coll_libnbc_ireduce(const void* sendbuf, void* recvbuf, int count, MPI_
   MPI_Aint ext;
   NBC_Schedule *schedule;
   char *redbuf=NULL, inplace;
+  void *tmpbuf;
   char tmpredbuf = 0;
   enum { NBC_RED_BINOMIAL, NBC_RED_CHAIN } alg;
-  NBC_Handle *handle;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
   ptrdiff_t span, gap;
 
@@ -90,11 +90,6 @@ int ompi_coll_libnbc_ireduce(const void* sendbuf, void* recvbuf, int count, MPI_
     return OMPI_SUCCESS;
   }
 
-  res = NBC_Init_handle (comm, &handle, libnbc_module);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    return res;
-  }
-
   span = opal_datatype_span(&datatype->super, count, &gap);
 
   /* algorithm selection */
@@ -102,23 +97,22 @@ int ompi_coll_libnbc_ireduce(const void* sendbuf, void* recvbuf, int count, MPI_
     alg = NBC_RED_BINOMIAL;
     if(rank == root) {
       /* root reduces in receivebuffer */
-      handle->tmpbuf = malloc (span);
+      tmpbuf = malloc (span);
       redbuf = recvbuf;
     } else {
       /* recvbuf may not be valid on non-root nodes */
       ptrdiff_t span_align = OPAL_ALIGN(span, datatype->super.align, ptrdiff_t);
-      handle->tmpbuf = malloc (span_align + span);
+      tmpbuf = malloc (span_align + span);
       redbuf = (char*)span_align - gap;
       tmpredbuf = 1;
     }
   } else {
-    handle->tmpbuf = malloc (span);
+    tmpbuf = malloc (span);
     alg = NBC_RED_CHAIN;
     segsize = 16384/2;
   }
 
-  if (OPAL_UNLIKELY(NULL == handle->tmpbuf)) {
-    NBC_Return_handle (handle);
+  if (OPAL_UNLIKELY(NULL == tmpbuf)) {
     return OMPI_ERR_OUT_OF_RESOURCE;
   }
 
@@ -137,30 +131,29 @@ int ompi_coll_libnbc_ireduce(const void* sendbuf, void* recvbuf, int count, MPI_
 #endif
     schedule = OBJ_NEW(NBC_Schedule);
     if (OPAL_UNLIKELY(NULL == schedule)) {
-      NBC_Return_handle (handle);
+      free(tmpbuf);
       return OMPI_ERR_OUT_OF_RESOURCE;
     }
 
-    /* make sure the schedule is released with the handle on error */
-    handle->schedule = schedule;
-
     switch(alg) {
       case NBC_RED_BINOMIAL:
-        res = red_sched_binomial(rank, p, root, sendbuf, redbuf, tmpredbuf, count, datatype, op, inplace, schedule, handle);
+        res = red_sched_binomial(rank, p, root, sendbuf, redbuf, tmpredbuf, count, datatype, op, inplace, schedule, tmpbuf);
         break;
       case NBC_RED_CHAIN:
-        res = red_sched_chain(rank, p, root, sendbuf, recvbuf, count, datatype, op, ext, size, schedule, handle, segsize);
+        res = red_sched_chain(rank, p, root, sendbuf, recvbuf, count, datatype, op, ext, size, schedule, tmpbuf, segsize);
         break;
     }
 
     if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-      NBC_Return_handle (handle);
+      OBJ_RELEASE(schedule);
+      free(tmpbuf);
       return res;
     }
 
     res = NBC_Sched_commit(schedule);
     if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-      NBC_Return_handle (handle);
+      OBJ_RELEASE(schedule);
+      free(tmpbuf);
       return res;
     }
 #ifdef NBC_CACHE_SCHEDULE
@@ -195,15 +188,13 @@ int ompi_coll_libnbc_ireduce(const void* sendbuf, void* recvbuf, int count, MPI_
   }
 #endif
 
-  res = NBC_Start(handle, schedule);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, tmpbuf);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
+    OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return res;
   }
 
-  *request = (ompi_request_t *) handle;
-
-  /* tmpbuf is freed with the handle */
   return OMPI_SUCCESS;
 }
 
@@ -212,52 +203,46 @@ int ompi_coll_libnbc_ireduce_inter(const void* sendbuf, void* recvbuf, int count
 				   struct mca_coll_base_module_2_1_0_t *module) {
   int rank, res, rsize;
   NBC_Schedule *schedule;
-  NBC_Handle *handle;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
   ptrdiff_t span, gap;
+  void *tmpbuf;
 
   rank = ompi_comm_rank (comm);
   rsize = ompi_comm_remote_size (comm);
 
-  res = NBC_Init_handle(comm, &handle, libnbc_module);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    return res;
-  }
-
   span = opal_datatype_span(&datatype->super, count, &gap);
-  handle->tmpbuf = malloc (span);
-  if (OPAL_UNLIKELY(NULL == handle->tmpbuf)) {
-    NBC_Return_handle (handle);
+  tmpbuf = malloc (span);
+  if (OPAL_UNLIKELY(NULL == tmpbuf)) {
     return OMPI_ERR_OUT_OF_RESOURCE;
   }
 
   schedule = OBJ_NEW(NBC_Schedule);
   if (OPAL_UNLIKELY(NULL == schedule)) {
-    NBC_Return_handle (handle);
+    free(tmpbuf);
     return OMPI_ERR_OUT_OF_RESOURCE;
   }
 
-  res = red_sched_linear (rank, rsize, root, sendbuf, recvbuf, (void *)(-gap), count, datatype, op, schedule, handle);
+  res = red_sched_linear (rank, rsize, root, sendbuf, recvbuf, (void *)(-gap), count, datatype, op, schedule);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
+    OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return OMPI_ERR_OUT_OF_RESOURCE;
   }
 
   res = NBC_Sched_commit(schedule);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
-    return OMPI_ERR_OUT_OF_RESOURCE;
+    OBJ_RELEASE(schedule);
+    free(tmpbuf);
+    return res;
   }
 
-  res = NBC_Start(handle, schedule);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, tmpbuf);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
+    OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return OMPI_ERR_OUT_OF_RESOURCE;
   }
 
-  *request = (ompi_request_t *) handle;
-
-  /* tmpbuf is freed with the handle */
   return OMPI_SUCCESS;
 }
 
@@ -294,9 +279,9 @@ int ompi_coll_libnbc_ireduce_inter(const void* sendbuf, void* recvbuf, int count
   if (vrank == root) rank = 0; \
 }
 static inline int red_sched_binomial (int rank, int p, int root, const void *sendbuf, void *redbuf, char tmpredbuf, int count, MPI_Datatype datatype,
-                                      MPI_Op op, char inplace, NBC_Schedule *schedule, NBC_Handle *handle) {
+                                      MPI_Op op, char inplace, NBC_Schedule *schedule, void *tmpbuf) {
   int vroot, vrank, vpeer, peer, res, maxr;
-  char *rbuf, *lbuf, *buf, tmpbuf;
+  char *rbuf, *lbuf, *buf;
   int tmprbuf, tmplbuf;
   ptrdiff_t gap;
   (void)opal_datatype_span(&datatype->super, count, &gap);
@@ -325,7 +310,7 @@ static inline int red_sched_binomial (int rank, int p, int root, const void *sen
     rbuf = redbuf;
     tmprbuf = tmpredbuf;
     if (inplace) {
-        res = NBC_Copy(rbuf, count, datatype, ((char *)handle->tmpbuf)-gap, count, datatype, MPI_COMM_SELF);
+        res = NBC_Copy(rbuf, count, datatype, ((char *)tmpbuf)-gap, count, datatype, MPI_COMM_SELF);
         if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
           return res;
         }
@@ -338,6 +323,7 @@ static inline int red_sched_binomial (int rank, int p, int root, const void *sen
       vpeer = vrank + (1 << (r - 1));
       VRANK2RANK(peer, vpeer, vroot)
       if (peer < p) {
+        int tbuf;
         /* we have to wait until we have the data */
         res = NBC_Sched_recv (rbuf, tmprbuf, count, datatype, peer, schedule, true);
         if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
@@ -345,7 +331,7 @@ static inline int red_sched_binomial (int rank, int p, int root, const void *sen
         }
 
         /* perform the reduce in my local buffer */
-        /* this cannot be done until handle->tmpbuf is unused :-( so barrier after the op */
+        /* this cannot be done until tmpbuf is unused :-( so barrier after the op */
         if (firstred && !inplace) {
           /* perform the reduce with the senbuf */
           res = NBC_Sched_op (sendbuf, false, rbuf, tmprbuf, count, datatype, op, schedule, true);
@@ -360,7 +346,7 @@ static inline int red_sched_binomial (int rank, int p, int root, const void *sen
         }
         /* swap left and right buffers */
         buf = rbuf; rbuf = lbuf ; lbuf = buf;
-        tmpbuf = tmprbuf; tmprbuf = tmplbuf; tmplbuf = tmpbuf;
+        tbuf = tmprbuf; tmprbuf = tmplbuf; tmplbuf = tbuf;
       }
     } else {
       /* we have to send this round */
@@ -396,7 +382,7 @@ static inline int red_sched_binomial (int rank, int p, int root, const void *sen
 
 /* chain send ... */
 static inline int red_sched_chain (int rank, int p, int root, const void *sendbuf, void *recvbuf, int count, MPI_Datatype datatype,
-                                   MPI_Op op, int ext, size_t size, NBC_Schedule *schedule, NBC_Handle *handle, int fragsize) {
+                                   MPI_Op op, int ext, size_t size, NBC_Schedule *schedule, void *tmpbuf, int fragsize) {
   int res, vrank, rpeer, speer, numfrag, fragcount, thiscount;
   long offset;
 
@@ -474,7 +460,7 @@ static inline int red_sched_chain (int rank, int p, int root, const void *sendbu
 
 /* simple linear algorithm for intercommunicators */
 static inline int red_sched_linear (int rank, int rsize, int root, const void *sendbuf, void *recvbuf, void *tmpbuf, int count, MPI_Datatype datatype,
-                                    MPI_Op op, NBC_Schedule *schedule, NBC_Handle *handle) {
+                                    MPI_Op op, NBC_Schedule *schedule) {
   int res;
   char *rbuf, *lbuf, *buf;
   int tmprbuf, tmplbuf;
diff --git a/ompi/mca/coll/libnbc/nbc_ireduce_scatter.c b/ompi/mca/coll/libnbc/nbc_ireduce_scatter.c
index 58ce5334191..b19f95626a8 100644
--- a/ompi/mca/coll/libnbc/nbc_ireduce_scatter.c
+++ b/ompi/mca/coll/libnbc/nbc_ireduce_scatter.c
@@ -7,7 +7,7 @@
  *                         rights reserved.
  * Copyright (c) 2013-2015 Los Alamos National Security, LLC. All rights
  *                         reserved.
- * Copyright (c) 2014-2016 Research Organization for Information Science
+ * Copyright (c) 2014-2017 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  * Copyright (c) 2015      The University of Tennessee and The University
  *                         of Tennessee Research Foundation. All rights
@@ -45,7 +45,7 @@ int ompi_coll_libnbc_ireduce_scatter(const void* sendbuf, void* recvbuf, const i
   ptrdiff_t gap, span, span_align;
   char *sbuf, inplace;
   NBC_Schedule *schedule;
-  NBC_Handle *handle;
+  void *tmpbuf;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
   char *rbuf, *lbuf, *buf;
 
@@ -78,18 +78,12 @@ int ompi_coll_libnbc_ireduce_scatter(const void* sendbuf, void* recvbuf, const i
     return OMPI_SUCCESS;
   }
 
-  res = NBC_Init_handle (comm, &handle, libnbc_module);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    return res;
-  }
-
   maxr = (int) ceil ((log((double) p) / LOG2));
 
   span = opal_datatype_span(&datatype->super, count, &gap);
   span_align = OPAL_ALIGN(span, datatype->super.align, ptrdiff_t);
-  handle->tmpbuf = malloc (span_align + span);
-  if (OPAL_UNLIKELY(NULL == handle->tmpbuf)) {
-    NBC_Return_handle (handle);
+  tmpbuf = malloc (span_align + span);
+  if (OPAL_UNLIKELY(NULL == tmpbuf)) {
     return OMPI_ERR_OUT_OF_RESOURCE;
   }
 
@@ -98,13 +92,10 @@ int ompi_coll_libnbc_ireduce_scatter(const void* sendbuf, void* recvbuf, const i
 
   schedule = OBJ_NEW(NBC_Schedule);
   if (OPAL_UNLIKELY(NULL == schedule)) {
-    NBC_Return_handle (handle);
+    free(tmpbuf);
     return OMPI_ERR_OUT_OF_RESOURCE;
   }
 
-  /* make sure the schedule is released with the handle on error */
-  handle->schedule = schedule;
-
   for (int r = 1, firstred = 1 ; r <= maxr ; ++r) {
     if ((rank % (1 << r)) == 0) {
       /* we have to receive this round */
@@ -113,11 +104,12 @@ int ompi_coll_libnbc_ireduce_scatter(const void* sendbuf, void* recvbuf, const i
         /* we have to wait until we have the data */
         res = NBC_Sched_recv(rbuf, true, count, datatype, peer, schedule, true);
         if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-          NBC_Return_handle (handle);
+          OBJ_RELEASE(schedule);
+          free(tmpbuf);
           return res;
         }
 
-        /* this cannot be done until handle->tmpbuf is unused :-( so barrier after the op */
+        /* this cannot be done until tmpbuf is unused :-( so barrier after the op */
         if (firstred) {
           /* take reduce data from the sendbuf in the first round -> save copy */
           res = NBC_Sched_op (sendbuf, false, rbuf, true, count, datatype, op, schedule, true);
@@ -128,7 +120,8 @@ int ompi_coll_libnbc_ireduce_scatter(const void* sendbuf, void* recvbuf, const i
         }
 
         if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-          NBC_Return_handle (handle);
+          OBJ_RELEASE(schedule);
+          free(tmpbuf);
           return res;
         }
         /* swap left and right buffers */
@@ -145,7 +138,8 @@ int ompi_coll_libnbc_ireduce_scatter(const void* sendbuf, void* recvbuf, const i
         res = NBC_Sched_send (lbuf, true, count, datatype, peer, schedule, false);
       }
       if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-        NBC_Return_handle (handle);
+        OBJ_RELEASE(schedule);
+        free(tmpbuf);
         return res;
       }
 
@@ -156,7 +150,8 @@ int ompi_coll_libnbc_ireduce_scatter(const void* sendbuf, void* recvbuf, const i
 
   res = NBC_Sched_barrier(schedule);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
+    OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return res;
   }
 
@@ -169,7 +164,8 @@ int ompi_coll_libnbc_ireduce_scatter(const void* sendbuf, void* recvbuf, const i
       res = NBC_Sched_send (sbuf, true, recvcounts[r], datatype, r, schedule,
                             false);
       if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-        NBC_Return_handle (handle);
+        OBJ_RELEASE(schedule);
+        free(tmpbuf);
         return res;
       }
     }
@@ -181,25 +177,25 @@ int ompi_coll_libnbc_ireduce_scatter(const void* sendbuf, void* recvbuf, const i
   }
 
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
+    OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return res;
   }
 
   res = NBC_Sched_commit (schedule);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
+    OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return res;
   }
 
-  res = NBC_Start (handle, schedule);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, tmpbuf);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
+    OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return res;
   }
 
-  *request = (ompi_request_t *) handle;
-
-  /* tmpbuf is freed with the handle */
   return OMPI_SUCCESS;
 }
 
@@ -210,7 +206,7 @@ int ompi_coll_libnbc_ireduce_scatter_inter (const void* sendbuf, void* recvbuf,
   MPI_Aint ext;
   ptrdiff_t gap, span, span_align;
   NBC_Schedule *schedule;
-  NBC_Handle *handle;
+  void *tmpbuf = NULL;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
 
   rank = ompi_comm_rank (comm);
@@ -231,32 +227,24 @@ int ompi_coll_libnbc_ireduce_scatter_inter (const void* sendbuf, void* recvbuf,
   span = opal_datatype_span(&datatype->super, count, &gap);
   span_align = OPAL_ALIGN(span, datatype->super.align, ptrdiff_t);
 
-  res = NBC_Init_handle(comm, &handle, libnbc_module);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    return res;
-  }
-
   if (count > 0) {
-    handle->tmpbuf = malloc (span_align + span);
-    if (OPAL_UNLIKELY(NULL == handle->tmpbuf)) {
-      NBC_Return_handle (handle);
+    tmpbuf = malloc (span_align + span);
+    if (OPAL_UNLIKELY(NULL == tmpbuf)) {
       return OMPI_ERR_OUT_OF_RESOURCE;
     }
   }
 
   schedule = OBJ_NEW(NBC_Schedule);
   if (OPAL_UNLIKELY(NULL == schedule)) {
-    NBC_Return_handle (handle);
+    free(tmpbuf);
     return OMPI_ERR_OUT_OF_RESOURCE;
   }
 
-  /* make sure the schedule is released with the handle on error */
-  handle->schedule = schedule;
-
   /* send my data to the remote root */
   res = NBC_Sched_send(sendbuf, false, count, datatype, 0, schedule, false);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
+    OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return res;
   }
 
@@ -266,7 +254,8 @@ int ompi_coll_libnbc_ireduce_scatter_inter (const void* sendbuf, void* recvbuf,
     rbuf = (char *)(span_align-gap);
     res = NBC_Sched_recv (lbuf, true, count, datatype, 0, schedule, true);
     if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-      NBC_Return_handle (handle);
+      OBJ_RELEASE(schedule);
+      free(tmpbuf);
       return res;
     }
 
@@ -274,14 +263,16 @@ int ompi_coll_libnbc_ireduce_scatter_inter (const void* sendbuf, void* recvbuf,
       char *tbuf;
       res = NBC_Sched_recv (rbuf, true, count, datatype, peer, schedule, true);
       if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-        NBC_Return_handle (handle);
+        OBJ_RELEASE(schedule);
+        free(tmpbuf);
         return res;
       }
 
       res = NBC_Sched_op (lbuf, true, rbuf, true, count, datatype,
                           op, schedule, true);
       if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-        NBC_Return_handle (handle);
+        OBJ_RELEASE(schedule);
+        free(tmpbuf);
         return res;
       }
       tbuf = lbuf; lbuf = rbuf; rbuf = tbuf;
@@ -291,14 +282,16 @@ int ompi_coll_libnbc_ireduce_scatter_inter (const void* sendbuf, void* recvbuf,
     res = NBC_Sched_copy (lbuf, true, recvcounts[0], datatype, recvbuf, false,
                           recvcounts[0], datatype, schedule, false);
     if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-      NBC_Return_handle (handle);
+      OBJ_RELEASE(schedule);
+      free(tmpbuf);
       return res;
     }
     for (int peer = 1, offset = recvcounts[0] * ext; peer < lsize ; ++peer) {
       res = NBC_Sched_local_send (lbuf + offset, true, recvcounts[peer], datatype, peer, schedule,
                                   false);
       if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-        NBC_Return_handle (handle);
+        OBJ_RELEASE(schedule);
+        free(tmpbuf);
         return res;
       }
 
@@ -308,25 +301,25 @@ int ompi_coll_libnbc_ireduce_scatter_inter (const void* sendbuf, void* recvbuf,
     /* receive my block */
     res = NBC_Sched_local_recv (recvbuf, false, recvcounts[rank], datatype, 0, schedule, false);
     if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-      NBC_Return_handle (handle);
+      OBJ_RELEASE(schedule);
+      free(tmpbuf);
       return res;
     }
   }
 
   res = NBC_Sched_commit (schedule);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
+    OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return res;
   }
 
-  res = NBC_Start (handle, schedule);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, tmpbuf);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
+    OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return res;
   }
 
-  *request = (ompi_request_t *) handle;
-
-  /* tmpbuf is freed with the handle */
   return OMPI_SUCCESS;
 }
diff --git a/ompi/mca/coll/libnbc/nbc_ireduce_scatter_block.c b/ompi/mca/coll/libnbc/nbc_ireduce_scatter_block.c
index f05c9d5cb09..aec2f79d211 100644
--- a/ompi/mca/coll/libnbc/nbc_ireduce_scatter_block.c
+++ b/ompi/mca/coll/libnbc/nbc_ireduce_scatter_block.c
@@ -8,7 +8,7 @@
  * Copyright (c) 2012      Sandia National Laboratories. All rights reserved.
  * Copyright (c) 2013-2015 Los Alamos National Security, LLC. All rights
  *                         reserved.
- * Copyright (c) 2014-2016 Research Organization for Information Science
+ * Copyright (c) 2014-2017 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  *
  * Author(s): Torsten Hoefler <htor@cs.indiana.edu>
@@ -43,7 +43,7 @@ int ompi_coll_libnbc_ireduce_scatter_block(const void* sendbuf, void* recvbuf, i
   ptrdiff_t gap, span;
   char *redbuf, *sbuf, inplace;
   NBC_Schedule *schedule;
-  NBC_Handle *handle;
+  void *tmpbuf = NULL;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
 
   NBC_IN_PLACE(sendbuf, recvbuf, inplace);
@@ -57,20 +57,11 @@ int ompi_coll_libnbc_ireduce_scatter_block(const void* sendbuf, void* recvbuf, i
     return (MPI_SUCCESS == res) ? MPI_ERR_SIZE : res;
   }
 
-  res = NBC_Init_handle(comm, &handle, libnbc_module);
-  if (OMPI_SUCCESS != res) {
-    return res;
-  }
-
   schedule = OBJ_NEW(NBC_Schedule);
   if (NULL == schedule) {
-    OMPI_COLL_LIBNBC_REQUEST_RETURN(handle);
     return OMPI_ERR_OUT_OF_RESOURCE;
   }
 
-  /* make sure the schedule is released with the handle on error */
-  handle->schedule = schedule;
-
   maxr = (int)ceil((log((double)p)/LOG2));
 
   count = p * recvcount;
@@ -81,23 +72,22 @@ int ompi_coll_libnbc_ireduce_scatter_block(const void* sendbuf, void* recvbuf, i
 
     span = opal_datatype_span(&datatype->super, count, &gap);
     span_align = OPAL_ALIGN(span, datatype->super.align, ptrdiff_t);
-    handle->tmpbuf = malloc (span_align + span);
-    if (NULL == handle->tmpbuf) {
-      OMPI_COLL_LIBNBC_REQUEST_RETURN(handle);
+    tmpbuf = malloc (span_align + span);
+    if (NULL == tmpbuf) {
       OBJ_RELEASE(schedule);
       return OMPI_ERR_OUT_OF_RESOURCE;
     }
 
     rbuf = (void *)(-gap);
     lbuf = (char *)(span_align - gap);
-    redbuf = (char *) handle->tmpbuf + span_align - gap;
+    redbuf = (char *) tmpbuf + span_align - gap;
 
     /* copy data to redbuf if we only have a single node */
     if ((p == 1) && !inplace) {
       res = NBC_Copy (sendbuf, count, datatype, redbuf, count, datatype, comm);
       if (OMPI_SUCCESS != res) {
-        NBC_Return_handle (handle);
         OBJ_RELEASE(schedule);
+        free(tmpbuf);
         return res;
       }
     }
@@ -110,7 +100,8 @@ int ompi_coll_libnbc_ireduce_scatter_block(const void* sendbuf, void* recvbuf, i
           /* we have to wait until we have the data */
           res = NBC_Sched_recv (rbuf, true, count, datatype, peer, schedule, true);
           if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-            NBC_Return_handle (handle);
+            OBJ_RELEASE(schedule);
+            free(tmpbuf);
             return res;
           }
 
@@ -124,7 +115,8 @@ int ompi_coll_libnbc_ireduce_scatter_block(const void* sendbuf, void* recvbuf, i
           }
 
           if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-            NBC_Return_handle (handle);
+            OBJ_RELEASE(schedule);
+            free(tmpbuf);
             return res;
           }
           /* swap left and right buffers */
@@ -142,7 +134,8 @@ int ompi_coll_libnbc_ireduce_scatter_block(const void* sendbuf, void* recvbuf, i
         }
 
         if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-          NBC_Return_handle (handle);
+          OBJ_RELEASE(schedule);
+          free(tmpbuf);
           return res;
         }
 
@@ -153,7 +146,8 @@ int ompi_coll_libnbc_ireduce_scatter_block(const void* sendbuf, void* recvbuf, i
 
     res = NBC_Sched_barrier(schedule);
     if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-      NBC_Return_handle (handle);
+      OBJ_RELEASE(schedule);
+      free(tmpbuf);
       return res;
     }
 
@@ -161,7 +155,8 @@ int ompi_coll_libnbc_ireduce_scatter_block(const void* sendbuf, void* recvbuf, i
     if (rank != 0) {
       res = NBC_Sched_recv (recvbuf, false, recvcount, datatype, 0, schedule, false);
       if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-        NBC_Return_handle (handle);
+        OBJ_RELEASE(schedule);
+        free(tmpbuf);
         return res;
       }
     } else {
@@ -171,7 +166,8 @@ int ompi_coll_libnbc_ireduce_scatter_block(const void* sendbuf, void* recvbuf, i
         /* root sends the right buffer to the right receiver */
         res = NBC_Sched_send (sbuf, true, recvcount, datatype, r, schedule, false);
         if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-          NBC_Return_handle (handle);
+          OBJ_RELEASE(schedule);
+          free(tmpbuf);
           return res;
         }
       }
@@ -181,7 +177,8 @@ int ompi_coll_libnbc_ireduce_scatter_block(const void* sendbuf, void* recvbuf, i
                               datatype, schedule, false);
       }
       if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-        NBC_Return_handle (handle);
+        OBJ_RELEASE(schedule);
+        free(tmpbuf);
         return res;
       }
     }
@@ -189,19 +186,18 @@ int ompi_coll_libnbc_ireduce_scatter_block(const void* sendbuf, void* recvbuf, i
 
   res = NBC_Sched_commit (schedule);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
+    OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return res;
   }
 
-  res = NBC_Start (handle, schedule);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, tmpbuf);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
+    OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return res;
   }
 
-  *request = (ompi_request_t *) handle;
-
-  /* tmpbuf is freed with the handle */
   return OMPI_SUCCESS;
 }
 
@@ -212,7 +208,7 @@ int ompi_coll_libnbc_ireduce_scatter_block_inter(const void *sendbuf, void *recv
   MPI_Aint ext;
   ptrdiff_t gap, span, span_align;
   NBC_Schedule *schedule;
-  NBC_Handle *handle;
+  void *tmpbuf = NULL;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
 
   rank = ompi_comm_rank (comm);
@@ -225,37 +221,29 @@ int ompi_coll_libnbc_ireduce_scatter_block_inter(const void *sendbuf, void *recv
     return res;
   }
 
-  res = NBC_Init_handle(comm, &handle, libnbc_module);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    return res;
-  }
-
   count = rcount * lsize;
 
   span = opal_datatype_span(&dtype->super, count, &gap);
   span_align = OPAL_ALIGN(span, dtype->super.align, ptrdiff_t);
 
   if (count > 0) {
-    handle->tmpbuf = malloc (span_align + span);
-    if (NULL == handle->tmpbuf) {
-      NBC_Return_handle (handle);
+    tmpbuf = malloc (span_align + span);
+    if (NULL == tmpbuf) {
       return OMPI_ERR_OUT_OF_RESOURCE;
     }
   }
 
   schedule = OBJ_NEW(NBC_Schedule);
   if (NULL == schedule) {
-    NBC_Return_handle (handle);
+    free(tmpbuf);
     return OMPI_ERR_OUT_OF_RESOURCE;
   }
 
-  /* make sure the schedule is released with the handle on error */
-  handle->schedule = schedule;
-
   /* send my data to the remote root */
   res = NBC_Sched_send (sendbuf, false, count, dtype, 0, schedule, false);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
+    OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return res;
   }
 
@@ -265,7 +253,8 @@ int ompi_coll_libnbc_ireduce_scatter_block_inter(const void *sendbuf, void *recv
     rbuf = (char *)(span_align-gap);
     res = NBC_Sched_recv (lbuf, true, count, dtype, 0, schedule, true);
     if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-      NBC_Return_handle (handle);
+      OBJ_RELEASE(schedule);
+      free(tmpbuf);
       return res;
     }
 
@@ -273,14 +262,16 @@ int ompi_coll_libnbc_ireduce_scatter_block_inter(const void *sendbuf, void *recv
       char *tbuf;
       res = NBC_Sched_recv (rbuf, true, count, dtype, peer, schedule, true);
       if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-        NBC_Return_handle (handle);
+        OBJ_RELEASE(schedule);
+        free(tmpbuf);
         return res;
       }
 
       res = NBC_Sched_op (lbuf, true, rbuf, true, count, dtype,
                           op, schedule, true);
       if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-        NBC_Return_handle (handle);
+        OBJ_RELEASE(schedule);
+        free(tmpbuf);
         return res;
       }
       tbuf = lbuf; lbuf = rbuf; rbuf = tbuf;
@@ -290,13 +281,15 @@ int ompi_coll_libnbc_ireduce_scatter_block_inter(const void *sendbuf, void *recv
     res = NBC_Sched_copy (lbuf, true, rcount, dtype, recvbuf, false, rcount,
                           dtype, schedule, false);
     if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-      NBC_Return_handle (handle);
+      OBJ_RELEASE(schedule);
+      free(tmpbuf);
       return res;
     }
     for (int peer = 1 ; peer < lsize ; ++peer) {
       res = NBC_Sched_local_send (lbuf + ext * rcount * peer, true, rcount, dtype, peer, schedule, false);
       if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-        NBC_Return_handle (handle);
+        OBJ_RELEASE(schedule);
+        free(tmpbuf);
         return res;
       }
     }
@@ -304,7 +297,8 @@ int ompi_coll_libnbc_ireduce_scatter_block_inter(const void *sendbuf, void *recv
     /* receive my block */
     res = NBC_Sched_local_recv(recvbuf, false, rcount, dtype, 0, schedule, false);
     if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-      NBC_Return_handle (handle);
+      OBJ_RELEASE(schedule);
+      free(tmpbuf);
       return res;
     }
   }
@@ -313,18 +307,17 @@ int ompi_coll_libnbc_ireduce_scatter_block_inter(const void *sendbuf, void *recv
 
   res = NBC_Sched_commit(schedule);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
+    OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return res;
   }
 
-  res = NBC_Start(handle, schedule);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, tmpbuf);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
+    OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return res;
   }
 
-  *request = (ompi_request_t *) handle;
-
-  /* tmpbuf is freed with the handle */
   return OMPI_SUCCESS;
 }
diff --git a/ompi/mca/coll/libnbc/nbc_iscan.c b/ompi/mca/coll/libnbc/nbc_iscan.c
index 5b8b0bbdc1d..2047ad8665e 100644
--- a/ompi/mca/coll/libnbc/nbc_iscan.c
+++ b/ompi/mca/coll/libnbc/nbc_iscan.c
@@ -5,7 +5,7 @@
  *                         Corporation.  All rights reserved.
  * Copyright (c) 2006      The Technical University of Chemnitz. All
  *                         rights reserved.
- * Copyright (c) 2014-2015 Research Organization for Information Science
+ * Copyright (c) 2014-2017 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  * Copyright (c) 2015      Los Alamos National Security, LLC. All rights
  *                         reserved.
@@ -47,8 +47,8 @@ int ompi_coll_libnbc_iscan(const void* sendbuf, void* recvbuf, int count, MPI_Da
   int rank, p, res;
   ptrdiff_t gap, span;
   NBC_Schedule *schedule;
+  void *tmpbuf = NULL;
   char inplace;
-  NBC_Handle *handle;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
 
   NBC_IN_PLACE(sendbuf, recvbuf, inplace);
@@ -64,11 +64,6 @@ int ompi_coll_libnbc_iscan(const void* sendbuf, void* recvbuf, int count, MPI_Da
     }
   }
 
-  res = NBC_Init_handle(comm, &handle, libnbc_module);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    return res;
-  }
-
 #ifdef NBC_CACHE_SCHEDULE
   NBC_Scan_args *args, *found, search;
 
@@ -83,34 +78,32 @@ int ompi_coll_libnbc_iscan(const void* sendbuf, void* recvbuf, int count, MPI_Da
 #endif
     schedule = OBJ_NEW(NBC_Schedule);
     if (OPAL_UNLIKELY(NULL == schedule)) {
-      NBC_Return_handle (handle);
       return OMPI_ERR_OUT_OF_RESOURCE;
     }
 
-    /* ensure the schedule is released with the handle */
-    handle->schedule = schedule;
-
     if(rank != 0) {
       span = opal_datatype_span(&datatype->super, count, &gap);
-      handle->tmpbuf = malloc (span);
-      if (NULL == handle->tmpbuf) {
-        NBC_Return_handle (handle);
+      tmpbuf = malloc (span);
+      if (NULL == tmpbuf) {
+        OBJ_RELEASE(schedule);
         return OMPI_ERR_OUT_OF_RESOURCE;
       }
 
       /* we have to wait until we have the data */
       res = NBC_Sched_recv ((void *)(-gap), true, count, datatype, rank-1, schedule, true);
       if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-        NBC_Return_handle (handle);
+        OBJ_RELEASE(schedule);
+        free(tmpbuf);
         return res;
       }
 
       /* perform the reduce in my local buffer */
-      /* this cannot be done until handle->tmpbuf is unused :-( so barrier after the op */
+      /* this cannot be done until tmpbuf is unused :-( so barrier after the op */
       res = NBC_Sched_op ((void *)(-gap), true, recvbuf, false, count, datatype, op, schedule,
                           true);
       if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-        NBC_Return_handle (handle);
+        OBJ_RELEASE(schedule);
+        free(tmpbuf);
         return res;
       }
     }
@@ -118,14 +111,16 @@ int ompi_coll_libnbc_iscan(const void* sendbuf, void* recvbuf, int count, MPI_Da
     if (rank != p-1) {
       res = NBC_Sched_send (recvbuf, false, count, datatype, rank+1, schedule, false);
       if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-        NBC_Return_handle (handle);
+        OBJ_RELEASE(schedule);
+        free(tmpbuf);
         return res;
       }
     }
 
     res = NBC_Sched_commit (schedule);
     if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-      NBC_Return_handle (handle);
+      OBJ_RELEASE(schedule);
+      free(tmpbuf);
       return res;
     }
 
@@ -160,14 +155,12 @@ int ompi_coll_libnbc_iscan(const void* sendbuf, void* recvbuf, int count, MPI_Da
   }
 #endif
 
-  res = NBC_Start(handle, schedule);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, tmpbuf);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
+    OBJ_RELEASE(schedule);
+    free(tmpbuf);
     return res;
   }
 
-  *request = (ompi_request_t *) handle;
-
-  /* tmpbuf is freed with the handle */
   return OMPI_SUCCESS;
 }
diff --git a/ompi/mca/coll/libnbc/nbc_iscatter.c b/ompi/mca/coll/libnbc/nbc_iscatter.c
index 20481ad057a..c2c90c15401 100644
--- a/ompi/mca/coll/libnbc/nbc_iscatter.c
+++ b/ompi/mca/coll/libnbc/nbc_iscatter.c
@@ -10,7 +10,7 @@
  * Copyright (c) 2013      The University of Tennessee and The University
  *                         of Tennessee Research Foundation.  All rights
  *                         reserved.
- * Copyright (c) 2014-2016 Research Organization for Information Science
+ * Copyright (c) 2014-2017 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  *
  * Author(s): Torsten Hoefler <htor@cs.indiana.edu>
@@ -48,7 +48,6 @@ int ompi_coll_libnbc_iscatter(const void* sendbuf, int sendcount, MPI_Datatype s
   MPI_Aint sndext = 0;
   NBC_Schedule *schedule;
   char *sbuf, inplace = 0;
-  NBC_Handle *handle;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
 
 
@@ -154,20 +153,12 @@ int ompi_coll_libnbc_iscatter(const void* sendbuf, int sendcount, MPI_Datatype s
   }
 #endif
 
-  res = NBC_Init_handle(comm, &handle, libnbc_module);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
     OBJ_RELEASE(schedule);
     return res;
   }
 
-  res = NBC_Start(handle, schedule);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
-    return res;
-  }
-
-  *request = (ompi_request_t *) handle;
-
   return OMPI_SUCCESS;
 }
 
@@ -179,7 +170,6 @@ int ompi_coll_libnbc_iscatter_inter(const void* sendbuf, int sendcount, MPI_Data
     MPI_Aint sndext;
     NBC_Schedule *schedule;
     char *sbuf;
-    NBC_Handle *handle;
     ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
 
     rsize = ompi_comm_remote_size (comm);
@@ -223,19 +213,11 @@ int ompi_coll_libnbc_iscatter_inter(const void* sendbuf, int sendcount, MPI_Data
         return res;
     }
 
-    res = NBC_Init_handle(comm, &handle, libnbc_module);
+    res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
     if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
         OBJ_RELEASE(schedule);
         return res;
     }
 
-    res = NBC_Start(handle, schedule);
-    if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-        NBC_Return_handle (handle);
-        return res;
-    }
-
-    *request = (ompi_request_t *) handle;
-
     return OMPI_SUCCESS;
 }
diff --git a/ompi/mca/coll/libnbc/nbc_iscatterv.c b/ompi/mca/coll/libnbc/nbc_iscatterv.c
index 14ad8a5336c..991e7b72fb5 100644
--- a/ompi/mca/coll/libnbc/nbc_iscatterv.c
+++ b/ompi/mca/coll/libnbc/nbc_iscatterv.c
@@ -10,7 +10,7 @@
  * Copyright (c) 2013      The University of Tennessee and The University
  *                         of Tennessee Research Foundation.  All rights
  *                         reserved.
- * Copyright (c) 2014-2016 Research Organization for Information Science
+ * Copyright (c) 2014-2017 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  *
  * Author(s): Torsten Hoefler <htor@cs.indiana.edu>
@@ -31,7 +31,6 @@ int ompi_coll_libnbc_iscatterv(const void* sendbuf, const int *sendcounts, const
   MPI_Aint sndext;
   NBC_Schedule *schedule;
   char *sbuf, inplace = 0;
-  NBC_Handle *handle;
   ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
 
   rank = ompi_comm_rank (comm);
@@ -89,20 +88,12 @@ int ompi_coll_libnbc_iscatterv(const void* sendbuf, const int *sendcounts, const
     return res;
   }
 
-  res = NBC_Init_handle (comm, &handle, libnbc_module);
+  res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
   if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
     OBJ_RELEASE(schedule);
     return res;
   }
 
-  res = NBC_Start (handle, schedule);
-  if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-    NBC_Return_handle (handle);
-    return res;
-  }
-
-  *request = (ompi_request_t *) handle;
-
   return OMPI_SUCCESS;
 }
 
@@ -114,7 +105,6 @@ int ompi_coll_libnbc_iscatterv_inter (const void* sendbuf, const int *sendcounts
     MPI_Aint sndext;
     NBC_Schedule *schedule;
     char *sbuf;
-    NBC_Handle *handle;
     ompi_coll_libnbc_module_t *libnbc_module = (ompi_coll_libnbc_module_t*) module;
 
     rsize = ompi_comm_remote_size (comm);
@@ -157,19 +147,11 @@ int ompi_coll_libnbc_iscatterv_inter (const void* sendbuf, const int *sendcounts
         return res;
     }
 
-    res = NBC_Init_handle(comm, &handle, libnbc_module);
+    res = NBC_Schedule_request(schedule, comm, libnbc_module, request, NULL);
     if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
         OBJ_RELEASE(schedule);
         return res;
     }
 
-    res = NBC_Start(handle, schedule);
-    if (OPAL_UNLIKELY(OMPI_SUCCESS != res)) {
-        NBC_Return_handle (handle);
-        return res;
-    }
-
-    *request = (ompi_request_t *) handle;
-
     return OMPI_SUCCESS;
 }