diff --git a/ompi/mca/coll/hierarch/Makefile.am b/ompi/mca/coll/hierarch/Makefile.am
index 99e5627694..1165ee7e17 100644
--- a/ompi/mca/coll/hierarch/Makefile.am
+++ b/ompi/mca/coll/hierarch/Makefile.am
@@ -45,6 +45,7 @@ libmca_coll_hierarch_la_LDFLAGS = -module -avoid-version
 sources = \
         coll_hierarch.h \
         coll_hierarch.c \
+        coll_hierarch_allreduce.c \
         coll_hierarch_barrier.c \
         coll_hierarch_bcast.c \
         coll_hierarch_component.c \
diff --git a/ompi/mca/coll/hierarch/coll_hierarch.c b/ompi/mca/coll/hierarch/coll_hierarch.c
index 64219272ed..6ba5b03a7c 100644
--- a/ompi/mca/coll/hierarch/coll_hierarch.c
+++ b/ompi/mca/coll/hierarch/coll_hierarch.c
@@ -69,22 +69,22 @@ static const mca_coll_base_module_1_0_0_t intra = {
   /* Collective function pointers */
   /* function pointers marked with NULL are not yet implemented
      and will use the functions provided in the basic module */
-  NULL,
-  NULL,
-  NULL,
-  NULL,
-  NULL,
-  NULL,
+  NULL, /* allgather */
+  NULL, /* allgatherv */
+  mca_coll_hierarch_allreduce_intra, 
+  NULL, /* alltoall */
+  NULL, /* alltoallv */
+  NULL, /* alltoallw */
   mca_coll_hierarch_barrier_intra,
   mca_coll_hierarch_bcast_intra,
-  NULL, 
-  NULL, 
-  NULL, 
+  NULL,  /* exscan */
+  NULL,  /* gather */
+  NULL,  /* gatherv */
   mca_coll_hierarch_reduce_intra,
-  NULL, 
-  NULL,
-  NULL, 
-  NULL
+  NULL,  /* reduce_scatter */
+  NULL,  /* scan */
+  NULL,  /* scatter */
+  NULL   /* scatterv */
 };
 
 
diff --git a/ompi/mca/coll/hierarch/coll_hierarch.h b/ompi/mca/coll/hierarch/coll_hierarch.h
index aaf00129e7..5bec28bc79 100644
--- a/ompi/mca/coll/hierarch/coll_hierarch.h
+++ b/ompi/mca/coll/hierarch/coll_hierarch.h
@@ -37,6 +37,7 @@ extern int mca_coll_hierarch_priority_param;
 extern int mca_coll_hierarch_verbose_param;
 extern int mca_coll_hierarch_use_rdma_param;
 extern int mca_coll_hierarch_ignore_sm_param;
+extern int mca_coll_hier_verbose;
 
 
 #define HIER_DEFAULT_NUM_LLEAD 5
diff --git a/ompi/mca/coll/hierarch/coll_hierarch_allreduce.c b/ompi/mca/coll/hierarch/coll_hierarch_allreduce.c
new file mode 100644
index 0000000000..4a6d3af236
--- /dev/null
+++ b/ompi/mca/coll/hierarch/coll_hierarch_allreduce.c
@@ -0,0 +1,105 @@
+/*
+ * Copyright (c) 2004-2005 The Trustees of Indiana University.
+ *                         All rights reserved.
+ * Copyright (c) 2004-2005 The Trustees of the University of Tennessee.
+ *                         All rights reserved.
+ * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart, 
+ *                         University of Stuttgart.  All rights reserved.
+ * Copyright (c) 2004-2005 The Regents of the University of California.
+ *                         All rights reserved.
+ * $COPYRIGHT$
+ * 
+ * Additional copyrights may follow
+ * 
+ * $HEADER$
+ */
+
+#include "ompi_config.h"
+#include "coll_hierarch.h"
+
+#include <stdio.h>
+
+#include "mpi.h"
+#include "ompi/include/constants.h"
+#include "opal/util/output.h"
+#include "communicator/communicator.h"
+#include "datatype/datatype.h"
+#include "mca/coll/coll.h"
+
+
+/*
+ *	reduce_intra
+ *
+ *	Function:	- reduction using two level hierarchy algorithm
+ *	Accepts:	- same as MPI_Reduce()
+ *	Returns:	- MPI_SUCCESS or error code
+ */
+int mca_coll_hierarch_allreduce_intra(void *sbuf, void *rbuf, int count,
+				      struct ompi_datatype_t *dtype, 
+				      struct ompi_op_t *op,
+				      struct ompi_communicator_t *comm)
+{
+    struct mca_coll_base_comm_t *data=NULL;
+    struct ompi_communicator_t *llcomm=NULL;
+    struct ompi_communicator_t *lcomm=NULL;
+    int rank, lrank;
+    int lroot, llroot;
+    long extent, true_extent, lb, true_lb;
+    char *tmpbuf=NULL, *tbuf=NULL;
+    int ret;
+    int root=0;
+
+    rank   = ompi_comm_rank ( comm );
+    data   = comm->c_coll_selected_data;
+    lcomm  = data->hier_lcomm;
+
+    if ( mca_coll_hier_verbose ) {
+      printf("%s:%d: executing hierarchical allreduce with cnt=%d \n",
+	     comm->c_name, rank, count );
+    }
+
+    llcomm = mca_coll_hierarch_get_llcomm ( root, data, &llroot, &lroot);
+
+    if ( MPI_COMM_NULL != lcomm ) {
+      lrank = ompi_comm_rank (lcomm);
+	if ( lrank == lroot ) {
+	    ompi_ddt_get_extent(dtype, &lb, &extent);
+	    ompi_ddt_get_true_extent(dtype, &true_lb, &true_extent);
+
+	    tbuf = malloc(true_extent + (count - 1) * extent);
+	    if (NULL == tbuf) {
+		return OMPI_ERR_OUT_OF_RESOURCE;
+	    }
+	    tmpbuf = tbuf - lb;
+	}
+
+	if ( MPI_IN_PLACE != sbuf ) {
+	    ret = lcomm->c_coll.coll_reduce (sbuf, tmpbuf, count, dtype, 
+					     op, lroot, lcomm);
+	}
+	else {
+            ret = lcomm->c_coll.coll_reduce (rbuf, tmpbuf, count, dtype, 
+					     op, lroot, lcomm);
+	}
+	if ( OMPI_SUCCESS != ret ) {
+	    goto exit;
+	}
+    }
+
+    if ( MPI_UNDEFINED != llroot ) {
+	ret = llcomm->c_coll.coll_allreduce (tmpbuf, rbuf, count, dtype,
+					     op, llcomm);
+    }
+
+    if ( MPI_COMM_NULL != lcomm ) {
+	ret = lcomm->c_coll.coll_bcast(rbuf, count, dtype, lroot, lcomm );
+    }
+
+
+ exit:
+    if ( NULL != tmpbuf ) {
+	free ( tmpbuf );
+    }
+
+    return ret;
+}
diff --git a/ompi/mca/coll/hierarch/coll_hierarch_barrier.c b/ompi/mca/coll/hierarch/coll_hierarch_barrier.c
index 7bde9d8a72..1b8312726a 100644
--- a/ompi/mca/coll/hierarch/coll_hierarch_barrier.c
+++ b/ompi/mca/coll/hierarch/coll_hierarch_barrier.c
@@ -14,28 +14,66 @@
  * $HEADER$
  */
 
+
 #include "ompi_config.h"
 #include "coll_hierarch.h"
 
 #include "mpi.h"
 #include "ompi/include/constants.h"
 #include "opal/util/output.h"
+#include "communicator/communicator.h"
 #include "mca/coll/coll.h"
-#include "mca/coll/base/base.h"
-#include "coll_hierarch.h"
 
 
 /*
  *	barrier_intra
  *
- *	Function:	- barrier using O(N) algorithm
+ *	Function:	- barrier using hierarchical algorithm
  *	Accepts:	- same as MPI_Barrier()
  *	Returns:	- MPI_SUCCESS or error code
  */
 int mca_coll_hierarch_barrier_intra(struct ompi_communicator_t *comm)
 {
-  opal_output_verbose(10, mca_coll_base_output, "In hierarch barrier_intra");
-  return MPI_SUCCESS;
+    struct mca_coll_base_comm_t *data=NULL;
+    struct ompi_communicator_t *llcomm=NULL;
+    struct ompi_communicator_t *lcomm=NULL;
+    int root=0;
+    int lroot, llroot;
+    int rank, ret;
+    
+    rank   = ompi_comm_rank ( comm );
+    data   = comm->c_coll_selected_data;
+    lcomm  = data->hier_lcomm;
+
+    if ( mca_coll_hier_verbose ) { 
+      printf("%s:%d: executing hierarchical barrier\n", comm->c_name, rank );
+    } 
+
+    llcomm = mca_coll_hierarch_get_llcomm ( root, data, &llroot, &lroot);
+
+    /* 
+     * Barrier consists of three steps:
+     * - barrier on the low-level communicators
+     * - barrier among the local leaders
+     * - barrier on the low-level communicators. This step is 
+     *   necessary to avoid that any non local leaders exit too early.
+     */
+    if ( MPI_COMM_NULL != lcomm ) {
+	ret = lcomm->c_coll.coll_barrier ( lcomm );
+        if ( OMPI_SUCCESS != ret ) {
+           return ret;
+	}
+    }
+
+    if ( MPI_UNDEFINED != llroot ) {
+	ret = llcomm->c_coll.coll_barrier ( llcomm );
+    }
+
+    if ( MPI_COMM_NULL != lcomm ) {
+	ret = lcomm->c_coll.coll_barrier ( lcomm );
+    }
+
+    return  ret;
 }
 
 
diff --git a/ompi/mca/coll/hierarch/coll_hierarch_bcast.c b/ompi/mca/coll/hierarch/coll_hierarch_bcast.c
index 23a7d29511..863ffce855 100644
--- a/ompi/mca/coll/hierarch/coll_hierarch_bcast.c
+++ b/ompi/mca/coll/hierarch/coll_hierarch_bcast.c
@@ -31,8 +31,6 @@
  *	Returns:	- MPI_SUCCESS or error code
  */
 
-extern int mca_coll_hier_verbose;
-
 
 int mca_coll_hierarch_bcast_intra(void *buff, 
 				  int count,
diff --git a/ompi/mca/coll/hierarch/coll_hierarch_reduce.c b/ompi/mca/coll/hierarch/coll_hierarch_reduce.c
index ad0f958c41..9a2124a8df 100644
--- a/ompi/mca/coll/hierarch/coll_hierarch_reduce.c
+++ b/ompi/mca/coll/hierarch/coll_hierarch_reduce.c
@@ -26,7 +26,6 @@
 #include "datatype/datatype.h"
 #include "mca/coll/coll.h"
 
-extern int mca_coll_hier_verbose;
 
 /*
  *	reduce_intra