diff --git a/ompi/mca/coll/hierarch/coll_hierarch.c b/ompi/mca/coll/hierarch/coll_hierarch.c
index 330789b96e..6ecc9279ea 100644
--- a/ompi/mca/coll/hierarch/coll_hierarch.c
+++ b/ompi/mca/coll/hierarch/coll_hierarch.c
@@ -28,8 +28,6 @@
 #include "mca/coll/coll.h"
 #include "mca/coll/base/base.h"
 #include "ompi/mca/coll/base/coll_tags.h"
-#include "coll_hierarch.h"
-
 
 #include "class/ompi_bitmap.h"
 #include "mca/bml/bml.h"
diff --git a/ompi/mca/coll/hierarch/coll_hierarch_bcast.c b/ompi/mca/coll/hierarch/coll_hierarch_bcast.c
index 82182ac134..f672e680ae 100644
--- a/ompi/mca/coll/hierarch/coll_hierarch_bcast.c
+++ b/ompi/mca/coll/hierarch/coll_hierarch_bcast.c
@@ -22,9 +22,6 @@
 #include "opal/util/output.h"
 #include "communicator/communicator.h"
 #include "mca/coll/coll.h"
-#include "mca/coll/base/base.h"
-#include "mca/coll/base/coll_tags.h"
-#include "coll_hierarch.h"
 
 /*
  *	bcast_intra
diff --git a/ompi/mca/coll/hierarch/coll_hierarch_reduce.c b/ompi/mca/coll/hierarch/coll_hierarch_reduce.c
index 204957aba5..a508750a91 100644
--- a/ompi/mca/coll/hierarch/coll_hierarch_reduce.c
+++ b/ompi/mca/coll/hierarch/coll_hierarch_reduce.c
@@ -22,15 +22,15 @@
 #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"
-#include "mca/coll/base/base.h"
-#include "coll_hierarch.h"
 
 
 /*
  *	reduce_intra
  *
- *	Function:	- reduction using O(N) algorithm
+ *	Function:	- reduction using two level hierarchy algorithm
  *	Accepts:	- same as MPI_Reduce()
  *	Returns:	- MPI_SUCCESS or error code
  */
@@ -39,5 +39,57 @@ int mca_coll_hierarch_reduce_intra(void *sbuf, void *rbuf, int count,
                                struct ompi_op_t *op,
                                int root, struct ompi_communicator_t *comm)
 {
-    return  OMPI_SUCCESS;
+    struct mca_coll_base_comm_t *data=NULL;
+    struct ompi_communicator_t *llcomm=NULL;
+    struct ompi_communicator_t *lcomm=NULL;
+    int rank, lrank;
+    int ret;
+    int lleader, llroot;
+    long extent, true_extent, lb, true_lb;
+    char *tmpbuf=NULL, *tbuf=NULL;
+
+    rank   = ompi_comm_rank ( comm );
+    data   = comm->c_coll_selected_data;
+    lcomm  = data->hier_lcomm;
+
+    if ( MPI_COMM_NULL != lcomm ) {
+	mca_coll_hierarch_get_lleader ( root, data, &lleader );
+	
+	lrank = ompi_comm_rank (lcomm);
+	if ( lrank == lleader ) {
+	    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, lleader, lcomm);
+	}
+	else {
+            ret = lcomm->c_coll.coll_reduce (rbuf, tmpbuf, count, dtype, 
+					     op, lleader, lcomm);
+	}
+	if ( OMPI_SUCCESS != ret ) {
+	    goto exit;
+	}
+    }
+
+    llcomm = mca_coll_hierarch_get_llcomm ( root, data, &llroot);
+    if ( MPI_UNDEFINED != llroot ) {
+	ret = llcomm->c_coll.coll_reduce (tmpbuf, rbuf, count, dtype,
+					  op, llroot, llcomm);
+    }
+
+ exit:
+    if ( NULL != tmpbuf ) {
+	free ( tmpbuf );
+    }
+
+    return ret;
 }