X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/4cf3e4e9a47b349606e27b77d407c6bf0594f330..deba0f7f0441a96f742af845cfceabf340d0f27f:/src/simix/smx_network.c

diff --git a/src/simix/smx_network.c b/src/simix/smx_network.c
index 93e52716fa..4987f8f20e 100644
--- a/src/simix/smx_network.c
+++ b/src/simix/smx_network.c
@@ -1,13 +1,15 @@
-/* 	$Id$	 */
-
-/* Copyright (c) 2009 Cristian Rosa.
-   All rights reserved.                                          */
+/* Copyright (c) 2009, 2010. The SimGrid Team.
+ * All rights reserved.                                                     */
 
 /* This program is free software; you can redistribute it and/or modify it
  * under the terms of the license (GNU LGPL) which comes with this package. */
 
 #include "private.h"
 #include "xbt/log.h"
+#include "xbt/dict.h"
+
+/* Pimple to get an histogram of message sizes in the simulation */
+xbt_dict_t msg_sizes = NULL;
 
 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_network, simix,
                                 "Logging specific to SIMIX (network)");
@@ -208,12 +210,15 @@ static inline void SIMIX_communication_start(smx_comm_t comm)
     comm->act = SIMIX_action_communicate(comm->src_proc->smx_host, 
                                          comm->dst_proc->smx_host, NULL, 
                                          comm->task_size, comm->rate);
+#ifdef HAVE_TRACING
+    TRACE_smx_action_communicate (comm->act, comm->src_proc);
+#endif
 
     /* If any of the process is suspend, create the action but stop its execution,
        it will be restarted when the sender process resume */
     if(SIMIX_process_is_suspended(comm->src_proc) || 
        SIMIX_process_is_suspended(comm->dst_proc)) {
-      SIMIX_action_set_priority(comm->act, 0);
+      SIMIX_action_suspend(comm->act);
     }
     
     /* Add the communication as user data of the action */
@@ -232,9 +237,9 @@ static inline void SIMIX_communication_start(smx_comm_t comm)
  *  \param timeout The max amount of time to wait for the communication to finish
  *
  *  Throws:
- *   - host_error if peer failed
- *   - timeout_error if communication reached the timeout specified
- *   - network_error if network failed or peer issued a timeout
+ *   - host_error if local peer failed
+ *   - timeout_error if communication reached the timeout specified (either because of local peer or remote peer)
+ *   - network_error if network failed or remote peer failed
  */
 static inline void SIMIX_communication_wait_for_completion(smx_comm_t comm, double timeout)
 {
@@ -285,7 +290,7 @@ static inline void SIMIX_communication_wait_for_completion(smx_comm_t comm, doub
   SIMIX_sem_release_forever(comm->sem);
   
   /* Check for errors other than timeouts (they are catched above) */
-  if(!SIMIX_host_get_state(SIMIX_host_self())){
+  if (!SIMIX_host_get_state(SIMIX_host_self())){
     if(comm->rdv)
       SIMIX_rdv_remove(comm->rdv, comm);
     SIMIX_communication_destroy(comm);
@@ -293,7 +298,18 @@ static inline void SIMIX_communication_wait_for_completion(smx_comm_t comm, doub
   } else if (SIMIX_action_get_state(comm->act) == SURF_ACTION_FAILED){
     SIMIX_communication_destroy(comm);
     THROW0(network_error, 0, "Link failure");
+  } else if (!SIMIX_host_get_state(SIMIX_process_get_host(comm->dst_proc)) ||
+      !SIMIX_host_get_state(SIMIX_process_get_host(comm->src_proc))) {
+    /* We test both src&dst because we dunno who we are today, and we already tested myself above.
+     *    So, at the end, we test the remote peer only
+     * Moreover, we have to test it because if the remote peer fails, the action comm->act is not done nor failed.
+     *    In that case, we got awaken by the little endless actions created in the SIMIX_sem_acquire(comm->sem)
+     *    at the beginning of this function. */
+    SIMIX_communication_destroy(comm);
+    THROW0(network_error, 0, "Remote peer failed");
+
   }
+  SIMIX_communication_destroy(comm);
 }
 
 /**
@@ -325,7 +341,7 @@ void SIMIX_network_set_copy_data_callback(void (*callback)(smx_comm_t, size_t))
 }
 
 void SIMIX_network_copy_pointer_callback(smx_comm_t comm, size_t buff_size) {
-  xbt_assert1((buff_size == sizeof(void*)), "Cannot copy %d bytes: must be sizeof(void*)",buff_size);
+  xbt_assert1((buff_size == sizeof(void*)), "Cannot copy %zu bytes: must be sizeof(void*)",buff_size);
   *(void**)(comm->dst_buff) = comm->src_buff;
 }
 
@@ -339,12 +355,19 @@ void SIMIX_network_copy_buffer_callback(smx_comm_t comm, size_t buff_size) {
  */
 void SIMIX_network_copy_data(smx_comm_t comm)
 {
+  size_t buff_size = comm->src_buff_size;
+
+  DEBUG6("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)",
+      comm,
+      comm->src_proc->smx_host->name, comm->src_buff,
+      comm->dst_proc->smx_host->name, comm->dst_buff,
+      buff_size);
+
   /* If there is no data to be copy then return */
   if(!comm->src_buff || !comm->dst_buff)
     return;
   
   /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
-  size_t buff_size = comm->src_buff_size;
   if (comm->dst_buff_size)
     buff_size = MIN(buff_size,*(comm->dst_buff_size));
   
@@ -354,15 +377,31 @@ void SIMIX_network_copy_data(smx_comm_t comm)
 
   if(buff_size == 0)
     return;
-  DEBUG6("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)",
-      comm,
-      comm->src_proc->smx_host->name, comm->src_buff,
-      comm->dst_proc->smx_host->name, comm->dst_buff,
-      buff_size);
-  if(SIMIX_network_copy_data_callback) {
-    SIMIX_network_copy_data_callback(comm, buff_size);
+  (*SIMIX_network_copy_data_callback)(comm, buff_size);
+
+  /* pimple to display the message sizes */
+  {
+    if (msg_sizes == NULL)
+      msg_sizes = xbt_dict_new();
+    uintptr_t casted_size = comm->task_size;
+    uintptr_t amount = xbt_dicti_get(msg_sizes, casted_size);
+    amount++;
+
+    xbt_dicti_set(msg_sizes,casted_size, amount);
   }
 }
+#include "xbt.h"
+/* pimple to display the message sizes */
+void SIMIX_message_sizes_output(const char *filename) {
+  FILE * out = fopen(filename,"w");
+  xbt_assert1(out,"Cannot open file %s",filename);
+  uintptr_t key,data;
+  xbt_dict_cursor_t cursor;
+  xbt_dict_foreach(msg_sizes,cursor,key,data) {
+    fprintf(out,"%zu %zu\n",key,data);
+  }
+  fclose(out);
+}
 
 /**
  *  \brief Return the user data associated to the communication
@@ -475,8 +514,6 @@ smx_comm_t SIMIX_network_irecv(smx_rdv_t rdv, void *dst_buff, size_t *dst_buff_s
 XBT_INLINE void SIMIX_network_wait(smx_comm_t comm, double timeout) {
   /* Wait for communication completion */
   SIMIX_communication_wait_for_completion(comm, timeout);
-
-  SIMIX_communication_destroy(comm);
 }
 
 /** @Returns whether the (asynchronous) communication is done yet or not */
@@ -519,6 +556,7 @@ unsigned int SIMIX_network_waitany(xbt_dynar_t comms) {
     SIMIX_communication_destroy(comm_finished);
     THROW0(network_error, 0, "Link failure");
   }
+  SIMIX_communication_destroy(comm_finished);
 
   return found_comm;
 }