diff --git a/src/tree/tree.c b/src/tree/tree.c
index 061007441c1058e43bc1710b6b7a0300897d8d20..060d07a671fbd133352dd4f959b2ab50710ac1eb 100644
--- a/src/tree/tree.c
+++ b/src/tree/tree.c
@@ -17,6 +17,7 @@
 #include <stdlib.h>
 #include <string.h>
 #include <omp.h>
+#include <sys/sysinfo.h>
 
 #define WRITE_NGBH
 #define WRITE_TOP_NODES
@@ -1289,11 +1290,42 @@ void convert_heap_idx_to_global(global_context_t* ctx, heap* H)
     }
 }
 
+void print_diagnositcs(global_context_t* ctx, int k)
+{
+    MPI_Comm shmcomm;
+    MPI_Comm_split_type(MPI_COMM_WORLD, MPI_COMM_TYPE_SHARED, 0,
+                        MPI_INFO_NULL, &shmcomm);
+    int shm_world_size;
+    MPI_Comm_size(shmcomm, &shm_world_size);
+    MPI_DB_PRINT("[INFO] Got %d ranks per node \n",shm_world_size); 
+    /* data */
+    float_t memory_use = (float_t)ctx -> local_n_points * ctx -> dims * sizeof(float_t);
+    memory_use += (float_t)sizeof(datapoint_info_t)* (float_t)(ctx -> local_n_points); 
+    /* ngbh */
+    memory_use += (float_t)sizeof(heap_node)*(float_t)k * (float_t)(ctx -> local_n_points); 
+    memory_use = memory_use / 1e9 * shm_world_size;
+
+    MPI_DB_PRINT("       Got ~%d points per node and %d ngbh per points\n", ctx -> local_n_points * shm_world_size, k); 
+    MPI_DB_PRINT("       Expected to use ~%.2lfGB of memory for each node, plus memory required to communicate ngbh\n", memory_use); 
+    struct sysinfo info;
+    sysinfo(&info);
+    
+    if(memory_use > 0.5 * (float_t)info.freeram / 1e9)
+        MPI_DB_PRINT("/!\\    Projected memory usage is more than half of the node memory, may go into troubles while communicating ngbh\n"); 
+
+    MPI_Barrier(ctx -> mpi_communicator);
+}
+
 
 void mpi_ngbh_search(global_context_t* ctx, datapoint_info_t* dp_info, top_kdtree_t* top_tree, kdtree_v2* local_tree, float_t* data, int k)
 {
 	/* local search */
-	MPI_DB_PRINT("[MASTER] Local ngbh search ");
+    /* print diagnostics */
+    print_diagnositcs(ctx, k);
+
+    
+    MPI_DB_PRINT("[MASTER] Local ngbh search ");
+    MPI_Barrier(ctx -> mpi_communicator);
     #pragma omp parallel for
 	for(int p = 0; p < ctx -> local_n_points; ++p)
 	{
@@ -1842,7 +1874,7 @@ void simulate_master_read_and_scatter(int dims, size_t n, global_context_t *ctx)
 		// std_g0163178_Me14_091_0000
         
         /* 10^6 points ca.*/
-		data = read_data_file(ctx,"../norm_data/std_LR_091_0001",MY_TRUE);
+		//data = read_data_file(ctx,"../norm_data/std_LR_091_0001",MY_TRUE);
 
         /* 10^7 ~ 8M points */
         
@@ -1853,7 +1885,7 @@ void simulate_master_read_and_scatter(int dims, size_t n, global_context_t *ctx)
 
         //
         //34 M
-		//data = read_data_file(ctx,"../norm_data/std_g1212639_091_0001",MY_TRUE);
+		data = read_data_file(ctx,"../norm_data/std_g1212639_091_0001",MY_TRUE);
         ctx->dims = 5;
 
 		// ctx -> n_points = 48*5*2000;