Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
Add a new Infiniband network model, based on the works of Jerome Vienne.
authorAugustin Degomme <augustin.degomme@imag.fr>
Thu, 16 Oct 2014 11:35:26 +0000 (13:35 +0200)
committerAugustin Degomme <augustin.degomme@imag.fr>
Thu, 16 Oct 2014 13:04:04 +0000 (15:04 +0200)
This computes penalties on communications to account for the slowdowns experimented on real systems, and modeled by jerome
check http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf for details

known issue: doesn't work yet with permanent receive mode, as some surf actions are not released properly

src/include/surf/surf.h
src/surf/network_ib.cpp
src/surf/network_ib.hpp
src/surf/surf_interface.cpp

index 1a9d56e..c9f19e8 100644 (file)
@@ -1008,6 +1008,16 @@ XBT_PUBLIC_DATA(surf_network_model_t) surf_network_model;
  */
 XBT_PUBLIC(void) surf_network_model_init_SMPI(void);
 
+/** \ingroup SURF_models
+ *  \brief Same as network model 'LagrangeVelho', only with different correction factors.
+ *
+ * This model impelments a variant of the contention model on Infinband networks based on
+ * the works of Jérôme Vienne : http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf
+ *
+ *  \see surf_workstation_model_init_IB()
+ */
+XBT_PUBLIC(void) surf_network_model_init_IB(void);
+
 /** \ingroup SURF_models
  *  \brief Initializes the platform with the network model 'LegrandVelho'
  *
index 29aeb32..5be0d3e 100644 (file)
@@ -3,3 +3,223 @@
 
 /* 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 "network_ib.hpp"
+#include "simgrid/sg_config.h"
+#include "maxmin_private.hpp"
+
+XBT_LOG_EXTERNAL_DEFAULT_CATEGORY(surf_network);
+
+double Bs=0.925;
+double Be=0.965;
+double ys=1.35;
+
+
+static void IB_create_host_callback(sg_platf_host_cbarg_t t){
+  
+  static int id=0;
+// pour t->id -> rajouter une nouvelle struct dans le dict, pour stocker les comms actives
+  if(((NetworkIBModel*)surf_network_model)->active_nodes==NULL)
+    ((NetworkIBModel*)surf_network_model)->active_nodes=xbt_dict_new();
+  
+  IBNode* act = new IBNode(id);
+
+  id++;
+  xbt_dict_set(((NetworkIBModel*)surf_network_model)->active_nodes, t->id, act, NULL);
+}
+
+static void IB_action_state_changed_callback(NetworkActionPtr action, e_surf_action_state_t statein, e_surf_action_state_t stateout){
+ if(statein!=SURF_ACTION_RUNNING|| stateout!=SURF_ACTION_DONE)
+    return;
+  std::pair<IBNode*,IBNode*> pair = ((NetworkIBModel*)surf_network_model)->active_comms[action];
+  XBT_VERB("action %p finished", action);
+ ((NetworkIBModel*)surf_network_model)->updateIBfactors(action, pair.first, pair.second, 1);
+
+  ((NetworkIBModel*)surf_network_model)->active_comms.erase(action);
+  
+}
+
+
+static void IB_action_init_callback(NetworkActionPtr action,RoutingEdgePtr src, RoutingEdgePtr dst, double size, double rate){
+  if(((NetworkIBModel*)surf_network_model)->active_nodes==NULL)
+    xbt_die("IB comm added, without any node connected !");
+  
+  IBNode* act_src= (IBNode*) xbt_dict_get_or_null(((NetworkIBModel*)surf_network_model)->active_nodes, src->getName());
+  if(act_src==NULL)
+    xbt_die("could not find src node active comms !");
+  //act_src->rate=rate;
+  
+  IBNode* act_dst= (IBNode*) xbt_dict_get_or_null(((NetworkIBModel*)surf_network_model)->active_nodes, dst->getName());
+  if(act_dst==NULL)
+    xbt_die("could not find dst node active comms !");  
+ // act_dst->rate=rate;
+  
+  ((NetworkIBModel*)surf_network_model)->active_comms[action]=make_pair<IBNode*,IBNode*>(act_src, act_dst);
+  //post the action in the second dist, to retrieve in the other callback
+  XBT_VERB("action %p init", action);
+
+  ((NetworkIBModel*)surf_network_model)->updateIBfactors(action, act_src, act_dst, 0);
+  
+}
+
+
+
+/*********
+ * Model *
+ *********/
+
+/************************************************************************/
+/* New model based on MPI contention model for Infiniband platforms */
+/************************************************************************/
+/* @Inproceedings{mescal_vienne_phd, */
+/*  author={Jérôme Vienne}, */
+/*  title={prédiction de performances d’applications de calcul haute performance sur réseau Infiniband}, */
+/*  address={Grenoble FRANCE}, */
+/*  month=june, */
+/*  year={2010} */
+/*  } */
+void surf_network_model_init_IB(void)
+{
+
+  if (surf_network_model)
+    return;
+  surf_network_model = new NetworkIBModel();
+  net_define_callbacks();
+  xbt_dynar_push(model_list, &surf_network_model);
+  surf_callback_connect(networkActionStateChangedCallbacks, IB_action_state_changed_callback);
+  surf_callback_connect(networkCommunicateCallbacks, IB_action_init_callback);
+
+  sg_platf_host_add_cb(IB_create_host_callback);
+  xbt_cfg_setdefault_double(_sg_cfg_set, "network/weight_S", 8775);
+}
+
+NetworkIBModel::NetworkIBModel()
+ : NetworkSmpiModel() {
+  m_haveGap=false;
+  active_nodes=NULL;
+}
+
+NetworkIBModel::~NetworkIBModel()
+{
+  xbt_dict_cursor_t cursor = NULL;
+  IBNode* instance = NULL;
+  char *name = NULL;
+  xbt_dict_foreach(active_nodes, cursor, name, instance)
+    delete instance;
+  xbt_dict_free(&active_nodes);
+}
+
+void NetworkIBModel::computeIBfactors(IBNode *root) {
+  double penalized_bw=0.0;
+  double num_comm_out = (double) root->ActiveCommsUp.size();
+  double max_penalty_out=0.0;
+  //first, compute all outbound penalties to get their max
+  for (std::vector<ActiveComm*>::iterator it= root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
+    double my_penalty_out = 1.0;
+
+    if(num_comm_out!=1){
+      if((*it)->destination->nbActiveCommsDown > 2)//number of comms sent to the receiving node
+       my_penalty_out = num_comm_out * Bs * ys;
+      else
+       my_penalty_out = num_comm_out * Bs;
+    }
+
+    max_penalty_out = max(max_penalty_out,my_penalty_out);
+  }
+
+  for (std::vector<ActiveComm*>::iterator it= root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
+
+    //compute inbound penalty
+    double my_penalty_in = 1.0;
+    int nb_comms = (*it)->destination->nbActiveCommsDown;//total number of incoming comms
+    if(nb_comms!=1)
+      my_penalty_in = ((*it)->destination->ActiveCommsDown)[root] //number of comm sent to dest by root node
+                     * Be 
+                     * (*it)->destination->ActiveCommsDown.size();//number of different nodes sending to dest
+    
+    double penalty=max(my_penalty_in,max_penalty_out);
+    
+    double rate_before_update = (*it)->action->getBound();
+    //save initial rate of the action
+    if((*it)->init_rate==-1) 
+      (*it)->init_rate= rate_before_update;
+    
+    penalized_bw= ! num_comm_out ? (*it)->init_rate : (*it)->init_rate /penalty;
+    
+    if (!double_equals(penalized_bw, rate_before_update, sg_surf_precision)){
+     // XBT_VERB("%d->%d action %p penalty updated : in %f, out %f, final %f, before %f , initial rate %f", root->id,(*it)->destination->id,(*it)->action,my_penalty_in,my_penalty_out,penalized_bw, (*it)->action->getBound(), (*it)->init_rate );
+      lmm_update_variable_bound(p_maxminSystem, (*it)->action->getVariable(), penalized_bw);
+    }/*else{
+     // XBT_VERB("%d->%d action %p penalty not updated : in %f, out %f, final %f, initial rate %f", root->id,(*it)->destination->id,(*it)->action,my_penalty_in,my_penalty_out,penalized_bw, (*it)->init_rate );
+    }*/
+
+  }
+  XBT_VERB("Finished computing");
+}
+
+void NetworkIBModel::updateIBfactors_rec(IBNode *root, bool* updatedlist) {
+  if(updatedlist[root->id]==0){
+    XBT_VERB("Updating rec %d", root->id);
+    computeIBfactors(root);
+    updatedlist[root->id]=1;
+    for (std::vector<ActiveComm*>::iterator it= root->ActiveCommsUp.begin(); it != root->ActiveCommsUp.end(); ++it) {
+        if(updatedlist[(*it)->destination->id]!=1)
+          updateIBfactors_rec((*it)->destination, updatedlist);
+    }
+    for (std::map<IBNode*, int>::iterator it= root->ActiveCommsDown.begin(); it != root->ActiveCommsDown.end(); ++it) {
+        if(updatedlist[it->first->id]!=1)
+          updateIBfactors_rec(it->first, updatedlist);
+    }
+  }
+}
+
+
+void NetworkIBModel::updateIBfactors(NetworkActionPtr action, IBNode *from, IBNode * to, int remove) {
+  if (from == to)//disregard local comms (should use loopback)
+    return;
+  
+  bool* updated=(bool*)xbt_malloc0(xbt_dict_size(active_nodes)*sizeof(bool));
+  ActiveComm* comm=NULL;
+  if(remove){
+    if(to->ActiveCommsDown[from]==1)
+      to->ActiveCommsDown.erase(from);
+    else
+      to->ActiveCommsDown[from]-=1;
+
+    to->nbActiveCommsDown--;
+    for (std::vector<ActiveComm*>::iterator it= from->ActiveCommsUp.begin(); 
+        it != from->ActiveCommsUp.end(); ++it) {
+      XBT_VERB("inside vector");
+      if((*it)->action==action){
+       comm=(*it);
+       from->ActiveCommsUp.erase(it);
+        XBT_VERB("action deleted");
+       break;
+      }
+    }
+    action->unref();
+    /*from->ActiveCommsUp.erase(
+        std::remove(from->ActiveCommsUp.begin(),
+            from->ActiveCommsUp.end(), make_pair(to, action)),
+            from->ActiveCommsUp.end());*/
+
+  }else{
+    //from->ActiveCommsUp.push_back(std::make_pair(to, action));
+    action->ref();
+
+    ActiveComm* comm=new ActiveComm();
+    comm->action=action;
+    comm->destination=to;
+    from->ActiveCommsUp.push_back(comm);
+
+    to->ActiveCommsDown[from]+=1;
+    to->nbActiveCommsDown++;
+  }
+  XBT_VERB("Updating %d", from->id);
+  updateIBfactors_rec(from, updated);
+  XBT_VERB("Finished updating %d", from->id);
+  if(comm)delete comm;
+  xbt_free(updated);
+}
index b7d1715..8adaf4a 100644 (file)
@@ -7,37 +7,46 @@
 #ifndef SURF_NETWORK_IB_HPP_
 #define SURF_NETWORK_IB_HPP_
 
-class NetworkIBModel : public NetworkModel {
-private:
-public:
-  NetworkIBModel();
-  NetworkIBModel(const char *name);
-  ~NetworkModel();
-  virtual ActionPtr communicate(RoutingEdgePtr src, RoutingEdgePtr dst,
-                                double size, double rate);
-  virtual NetworkLinkPtr createNetworkLink(const char *name,
-                                          double bw_initial,
-                                          tmgr_trace_t bw_trace,
-                                          double lat_initial,
-                                          tmgr_trace_t lat_trace,
-                                          e_surf_resource_state_t state_initial,
-                                          tmgr_trace_t state_trace,
-                                          e_surf_link_sharing_policy_t policy,
-                                          xbt_dict_t properties);
+#include "network_smpi.hpp"
+class IBNode;
+
+
+class ActiveComm{
+public :
+  //IBNode* origin;
+  IBNode* destination;
+  NetworkActionPtr action;
+  double init_rate;
+  ActiveComm() : destination(NULL),action(NULL),init_rate(-1){};
+  ~ActiveComm(){};
 };
 
-class NetworkIBLink : public NetworkLink {
+class IBNode{
+public :
+  int id;
+    //store related links, to ease computation of the penalties
+  std::vector<ActiveComm*> ActiveCommsUp;
+  //store the number of comms received from each node
+  std::map<IBNode*, int> ActiveCommsDown;
+  //number of comms the node is receiving
+  int nbActiveCommsDown;
+  IBNode(int id) : id(id),nbActiveCommsDown(0){};
+  ~IBNode(){};
+};
+
+class NetworkIBModel : public NetworkSmpiModel {
 private:
+  void updateIBfactors_rec(IBNode *root, bool* updatedlist);
+  void computeIBfactors(IBNode *root);
 public:
-  NetworkIBLink(NetworkModelPtr model, const char *name, xbt_dict_t props);
-  NetworkIBLink(NetworkModelPtr model, const char *name, xbt_dict_t props,
-                lmm_constraint_t constraint,
-                tmgr_history_t history,
-                tmgr_trace_t state_trace);
-  ~NetworkIBLink();
-  virtual void updateLatency(double value, double date=surf_get_clock());
-  virtual void updateBandwidth(double value, double date=surf_get_clock());
-
+  NetworkIBModel();
+  NetworkIBModel(const char *name);
+  ~NetworkIBModel();
+  void updateIBfactors(NetworkActionPtr action, IBNode *from, IBNode * to, int remove);
+  
+  xbt_dict_t active_nodes;
+  std::map<NetworkActionPtr , std::pair<IBNode*,IBNode*> > active_comms;
 
 };
+
 #endif
index 6de2b62..2e79954 100644 (file)
@@ -117,6 +117,9 @@ s_surf_model_description_t surf_network_model_description[] = {
   {"SMPI",
    "Realistic network model specifically tailored for HPC settings (accurate modeling of slow start with correction factors on three intervals: < 1KiB, < 64 KiB, >= 64 KiB)",
    surf_network_model_init_SMPI},
+  {"IB",
+   "Realistic network model specifically tailored for HPC settings, with Infiniband contention model",
+   surf_network_model_init_IB},
   {"CM02",
    "Legacy network analytic model (Very similar to LV08, but without corrective factors. The timings of small messages are thus poorly modeled).",
    surf_network_model_init_CM02},
@@ -872,7 +875,7 @@ void Action::setBound(double bound)
 {
   XBT_IN("(%p,%g)", this, bound);
   if (p_variable)
-    lmm_update_variable_bound(getModel()->getMaxminSystem(), getVariable(), bound);
+    lmm_update_variable_bound(getModel()->getMaxminSystem(), p_variable, bound);
 
   if (getModel()->getUpdateMechanism() == UM_LAZY && getLastUpdate()!=surf_get_clock())
     heapRemove(getModel()->getActionHeap());