Update link to SimGrid doc

[loba.git] / process.h

diff --git a/process.h b/process.h
index cc10b0494ba0710994be865891af6b153acba24b..55ee792e1e59d56751fbe6157e858ee1a0acde58 100644 (file)
--- a/process.h
+++ b/process.h
@@ -5,26 +5,30 @@
 //#undef USE_UNORDERED_MAP
 
 #include <algorithm>
 //#undef USE_UNORDERED_MAP
 
 #include <algorithm>

-#include <tr1/functional>
+#include <atomic>

 #ifdef USE_UNORDERED_MAP
 #ifdef USE_UNORDERED_MAP

-#  include <tr1/unordered_map>
-#  define MAP_TEMPLATE std::tr1::unordered_map
+#  include <unordered_map>
+#  define MAP_TEMPLATE std::unordered_map

 #else
 #  include <map>
 #  define MAP_TEMPLATE std::map
 #endif
 #include <vector>
 #else
 #  include <map>
 #  define MAP_TEMPLATE std::map
 #endif
 #include <vector>

-#include <msg/msg.h>
+#include <simgrid/msg.h>

 #include <xbt/log.h>
 #include "communicator.h"
 #include "misc.h"
 #include "msg_thread.h"
 #include "neighbor.h"
 #include "options.h"
 #include <xbt/log.h>
 #include "communicator.h"
 #include "misc.h"
 #include "msg_thread.h"
 #include "neighbor.h"
 #include "options.h"

+#include "simgrid_features.h"

 #include "synchro.h"
 
 class process {
 public:
 #include "synchro.h"
 
 class process {
 public:

+    static void set_proc_mutex(mutex_t* m) { proc_mutex = m;            }
+
+    // Note: normally used with proc_mutex locked.

     static double get_total_load_init()    { return total_load_init;    }
     static double get_total_load_running() { return total_load_running; }
     static double get_total_load_exit()    { return total_load_exit;    }
     static double get_total_load_init()    { return total_load_init;    }
     static double get_total_load_running() { return total_load_running; }
     static double get_total_load_exit()    { return total_load_exit;    }


@@ -32,8 +36,21 @@ public:
     process(int argc, char* argv[]);
     virtual ~process();
 
     process(int argc, char* argv[]);
     virtual ~process();
 

-    double get_comp() const                { return comp; }
-    double get_real_load() const           { return real_load; }
+    double get_real_load() const           { return real_load;            }
+    double get_comp_amount() const         { return acc.comp_amount;      }
+    unsigned get_comp_iter() const         { return comp_iter;            }
+    unsigned get_all_comp_iter() const     { return all_comp_iter;        }
+    double get_iter_deviation() const;
+    double get_data_send_amount() const    { return acc.data_send.amount; }
+    double get_data_recv_amount() const    { return acc.data_recv.amount; }
+    unsigned get_data_send_count() const   { return acc.data_send.count;  }
+    unsigned get_data_recv_count() const   { return acc.data_recv.count;  }
+    double get_ctrl_send_amount() const    { return acc.ctrl_send.amount; }
+    double get_ctrl_recv_amount() const    { return acc.ctrl_recv.amount; }
+    unsigned get_ctrl_send_count() const   { return acc.ctrl_send.count;  }
+    unsigned get_ctrl_recv_count() const   { return acc.ctrl_recv.count;  }
+    double get_idle_duration() const       { return idle_duration;        }
+    double get_convergence() const         { return convergence;          }

 
     int run();
 
 
     int run();
 


@@ -70,14 +87,25 @@ protected:
                        xbt_log_category_t cat = _XBT_LOGV(default)) const;
 
 private:
                        xbt_log_category_t cat = _XBT_LOGV(default)) const;
 
 private:

+    static mutex_t *proc_mutex; // protect access to global variables
+                                // (must be set before constructing
+                                // the first object!)
+

     static double total_load_init; // sum of process loads at init
     static double total_load_running; // sum of loads while running
     static double total_load_exit; // sum of process loads at exit
 
     static double total_load_init; // sum of process loads at init
     static double total_load_running; // sum of loads while running
     static double total_load_exit; // sum of process loads at exit
 

-    typedef MAP_TEMPLATE<m_host_t, neighbor*> rev_neigh_type;
+    static int process_counter;
+    static double total_load_average;
+    static double average_load_ratio;
+    static double load_diff_threshold;
+
+    static std::atomic<int> convergence_counter;
+
+    typedef MAP_TEMPLATE<msg_host_t, neighbor*> rev_neigh_type;

     neigh_type neigh;           // list of neighbors (do not alter
                                 // after construction!)
     neigh_type neigh;           // list of neighbors (do not alter
                                 // after construction!)

-    rev_neigh_type rev_neigh;   // map m_host_t -> neighbor
+    rev_neigh_type rev_neigh;   // map msg_host_t -> neighbor

 
     communicator comm;          // communicator for this process
     int ctrl_close_pending;     // number of "close" messages to wait
 
     communicator comm;          // communicator for this process
     int ctrl_close_pending;     // number of "close" messages to wait


@@ -89,17 +117,57 @@ private:
 
     unsigned lb_iter;           // counter of load-balancing iterations
     unsigned comp_iter;         // counter of computation iterations
 
     unsigned lb_iter;           // counter of load-balancing iterations
     unsigned comp_iter;         // counter of computation iterations

-
-    double comp;                // total computing done so far (flops)
+    unsigned all_comp_iter;     // counter of computation iterations
+                                // (counting empty iterations too)

 
     double prev_load_broadcast; // used to ensure that we do not send
                                 // a same information messages
     double real_load;           // current load
     double expected_load;       // expected load in bookkeeping mode
 
     double prev_load_broadcast; // used to ensure that we do not send
                                 // a same information messages
     double real_load;           // current load
     double expected_load;       // expected load in bookkeeping mode

+    double received_load;       // load received from neighbors
+
+    double idle_duration;       // how long we had nothing to compute
+    double convergence;         // date when convergence was achieved, or -1.0
+    int local_convergence_counter; // number of iterations since convergence

 
     mutex_t mutex;              // synchronization between threads
     condition_t cond;
 
 
     mutex_t mutex;              // synchronization between threads
     condition_t cond;
 

+    struct mesg_accounting {
+        double amount;          // sum of message size
+        unsigned count;         // number of messages
+        mesg_accounting(): amount(0.0), count(0) { }
+    };
+    struct accounting {
+        double comp_amount;        // total computing done so far (flops)
+        mesg_accounting data_send; // data messages sent
+        mesg_accounting data_recv; // data messages received
+        mesg_accounting ctrl_send; // ctrl message sent
+        mesg_accounting ctrl_recv; // ctrl message received
+        accounting(): comp_amount(0.0) { }
+    };
+    accounting acc;             // use a structure so that it is
+                                // automatically initialized a
+                                // construction
+
+    void add_comp_amount(double amount) { acc.comp_amount += amount; }
+    void add_data_send_mesg(double amount) {
+        ++acc.data_send.count;
+        acc.data_send.amount += amount;
+    }
+    void add_data_recv_mesg(double amount) {
+        ++acc.data_recv.count;
+        acc.data_recv.amount += amount;
+    }
+    void add_ctrl_send_mesg(double amount) {
+        ++acc.ctrl_send.count;
+        acc.ctrl_send.amount += amount;
+    }
+    void add_ctrl_recv_mesg(double amount) {
+        ++acc.ctrl_recv.count;
+        acc.ctrl_recv.amount += amount;
+    }
+

     // Load-balancing loop
     msg_thread* lb_thread;
     void load_balance_loop();
     // Load-balancing loop
     msg_thread* lb_thread;
     void load_balance_loop();


@@ -107,12 +175,18 @@ private:
     // Simulate computation loop
     void compute_loop();
 
     // Simulate computation loop
     void compute_loop();
 

+    // Check for convergence
+    void convergence_check();
+

     // Check if we need to stop
     bool still_running();
 
     // Returns the sum of "to_send" for all neighbors.
     double get_sum_of_to_send() const;
 
     // Check if we need to stop
     bool still_running();
 
     // Returns the sum of "to_send" for all neighbors.
     double get_sum_of_to_send() const;
 

+    // Compute load_to_send (for data_send), subject to the execution parameters
+    static double compute_load_to_send(double desired);
+

     // Send procedures
     void ctrl_send(neighbor& nb);
     void data_send(neighbor& nb);
     // Send procedures
     void ctrl_send(neighbor& nb);
     void data_send(neighbor& nb);


@@ -124,19 +198,16 @@ private:
     // infinite waiting, or any positive timeout.
     void ctrl_receive(double timeout);
     void data_receive(double timeout);
     // infinite waiting, or any positive timeout.
     void ctrl_receive(double timeout);
     void data_receive(double timeout);

-    void handle_message(message* msg, m_host_t from);
+    void handle_message(message* msg, msg_host_t from);

 };
 
 template <typename Compare>
 void process::pneigh_sort_by_load(const Compare& comp)
 {
 };
 
 template <typename Compare>
 void process::pneigh_sort_by_load(const Compare& comp)
 {

-    using std::tr1::bind;
-    using std::tr1::placeholders::_1;
-    using std::tr1::placeholders::_2;

     std::sort(pneigh.begin(), pneigh.end(),
     std::sort(pneigh.begin(), pneigh.end(),

-              bind(comp,
-                   bind(&neighbor::get_load, _1),
-                   bind(&neighbor::get_load, _2)));
+              [&comp](const neighbor* a, const neighbor* b) {
+                  return comp(a->get_load(), b->get_load());
+              });

 }
 
 #endif // !PROCESS_H
 }
 
 #endif // !PROCESS_H