process.h

   1 #ifndef PROCESS_H
   2 #define PROCESS_H
   3
   4 #define USE_UNORDERED_MAP 1
   5 //#undef USE_UNORDERED_MAP
   6
   7 #include <algorithm>
   8 #include <functional>
   9 #ifdef USE_UNORDERED_MAP
  10 #  include <unordered_map>
  11 #  define MAP_TEMPLATE std::unordered_map
  12 #else
  13 #  include <map>
  14 #  define MAP_TEMPLATE std::map
  15 #endif
  16 #include <vector>
  17 #include <msg/msg.h>
  18 #include <xbt/log.h>
  19 #include "communicator.h"
  20 #include "misc.h"
  21 #include "msg_thread.h"
  22 #include "neighbor.h"
  23 #include "options.h"
  24 #include "synchro.h"
  25
  26 class process {
  27 public:
  28     static double get_total_load_init()    { return total_load_init;    }
  29     static double get_total_load_running() { return total_load_running; }
  30     static double get_total_load_exit()    { return total_load_exit;    }
  31
  32     process(int argc, char* argv[]);
  33     virtual ~process();
  34
  35     double get_real_load() const           { return real_load;            }
  36     double get_comp_amount() const         { return acc.comp_amount;      }
  37     unsigned get_comp_iter() const         { return comp_iter;            }
  38     unsigned get_all_comp_iter() const     { return all_comp_iter;        }
  39     double get_iter_deviation() const;
  40     double get_data_send_amount() const    { return acc.data_send.amount; }
  41     double get_data_recv_amount() const    { return acc.data_recv.amount; }
  42     unsigned get_data_send_count() const   { return acc.data_send.count;  }
  43     unsigned get_data_recv_count() const   { return acc.data_recv.count;  }
  44     double get_ctrl_send_amount() const    { return acc.ctrl_send.amount; }
  45     double get_ctrl_recv_amount() const    { return acc.ctrl_recv.amount; }
  46     unsigned get_ctrl_send_count() const   { return acc.ctrl_send.count;  }
  47     unsigned get_ctrl_recv_count() const   { return acc.ctrl_recv.count;  }
  48     double get_idle_duration() const       { return idle_duration;        }
  49     double get_convergence() const         { return convergence;          }
  50
  51     int run();
  52
  53 protected:
  54     typedef std::vector<neighbor> neigh_type;
  55     typedef std::vector<neighbor*> pneigh_type;
  56
  57     pneigh_type pneigh;         // list of pointers to neighbors that
  58                                 // we are free to reorder
  59
  60     // Get and set current load, which may be real load, or expected
  61     // load if opt::bookkeeping is true.
  62     double get_load() const                { return expected_load; }
  63
  64     // The load balancing algorithm comes here...
  65     virtual void load_balance();
  66
  67     // Register some amount of load to send to given neighbor.
  68     void send(neighbor& nb, double amount);
  69     void send(neighbor* nb, double amount) { send(*nb, amount); }
  70
  71     // Sort pneigh by applying comp to their loads
  72     template <typename Compare>
  73     void pneigh_sort_by_load(const Compare& comp);
  74
  75     // Calls neighbor::print(verbose, logp, cat) for each member of neigh.
  76     void print_loads(bool verbose = false,
  77                      e_xbt_log_priority_t logp = xbt_log_priority_info,
  78                      xbt_log_category_t cat = _XBT_LOGV(default)) const;
  79
  80     // Calls neighbor::print(verbose, logp, cat) for each member of pneigh.
  81     void print_loads_p(bool verbose = false,
  82                        e_xbt_log_priority_t logp = xbt_log_priority_info,
  83                        xbt_log_category_t cat = _XBT_LOGV(default)) const;
  84
  85 private:
  86     static double total_load_init; // sum of process loads at init
  87     static double total_load_running; // sum of loads while running
  88     static double total_load_exit; // sum of process loads at exit
  89
  90     static int process_counter;
  91     static double total_load_average;
  92     static double load_diff_threshold;
  93
  94     typedef MAP_TEMPLATE<m_host_t, neighbor*> rev_neigh_type;
  95     neigh_type neigh;           // list of neighbors (do not alter
  96                                 // after construction!)
  97     rev_neigh_type rev_neigh;   // map m_host_t -> neighbor
  98
  99     communicator comm;          // communicator for this process
 100     int ctrl_close_pending;     // number of "close" messages to wait
 101                                 // on ctrl channel
 102     int data_close_pending;     // number of "close" messages to wait
 103                                 // on data channel
 104     bool close_received;        // true if we received a "close" message
 105     bool finalizing;            // true when finalize() is running
 106
 107     unsigned lb_iter;           // counter of load-balancing iterations
 108     unsigned comp_iter;         // counter of computation iterations
 109     unsigned all_comp_iter;     // counter of computation iterations
 110                                 // (counting empty iterations too)
 111
 112     double prev_load_broadcast; // used to ensure that we do not send
 113                                 // a same information messages
 114     double real_load;           // current load
 115     double expected_load;       // expected load in bookkeeping mode
 116     double received_load;       // load received from neighbors
 117
 118     double idle_duration;       // how long we had nothing to compute
 119     double convergence;         // date when convergence was achieved, or -1.0
 120
 121     mutex_t mutex;              // synchronization between threads
 122     condition_t cond;
 123
 124     struct mesg_accounting {
 125         double amount;          // sum of message size
 126         unsigned count;         // number of messages
 127         mesg_accounting(): amount(0.0), count(0) { }
 128     };
 129     struct accounting {
 130         double comp_amount;        // total computing done so far (flops)
 131         mesg_accounting data_send; // data messages sent
 132         mesg_accounting data_recv; // data messages received
 133         mesg_accounting ctrl_send; // ctrl message sent
 134         mesg_accounting ctrl_recv; // ctrl message received
 135         accounting(): comp_amount(0.0) { }
 136     };
 137     accounting acc;             // use a structure so that it is
 138                                 // automatically initialized a
 139                                 // construction
 140
 141     void add_comp_amount(double amount) { acc.comp_amount += amount; }
 142     void add_data_send_mesg(double amount) {
 143         ++acc.data_send.count;
 144         acc.data_send.amount += amount;
 145     }
 146     void add_data_recv_mesg(double amount) {
 147         ++acc.data_recv.count;
 148         acc.data_recv.amount += amount;
 149     }
 150     void add_ctrl_send_mesg(double amount) {
 151         ++acc.ctrl_send.count;
 152         acc.ctrl_send.amount += amount;
 153     }
 154     void add_ctrl_recv_mesg(double amount) {
 155         ++acc.ctrl_recv.count;
 156         acc.ctrl_recv.amount += amount;
 157     }
 158
 159     // Load-balancing loop
 160     msg_thread* lb_thread;
 161     void load_balance_loop();
 162
 163     // Simulate computation loop
 164     void compute_loop();
 165
 166     // Check for convergence
 167     void convergence_check();
 168
 169     // Check if we need to stop
 170     bool still_running();
 171
 172     // Returns the sum of "to_send" for all neighbors.
 173     double get_sum_of_to_send() const;
 174
 175     // Compute load_to_send (for data_send), subject to the execution parameters
 176     static double compute_load_to_send(double desired);
 177
 178     // Send procedures
 179     void ctrl_send(neighbor& nb);
 180     void data_send(neighbor& nb);
 181     void ctrl_close(neighbor& nb);
 182     void data_close(neighbor& nb);
 183
 184     // Receive procedure
 185     // Parameter "timeout" may be 0 for non-blocking operation, -1 for
 186     // infinite waiting, or any positive timeout.
 187     void ctrl_receive(double timeout);
 188     void data_receive(double timeout);
 189     void handle_message(message* msg, m_host_t from);
 190 };
 191
 192 template <typename Compare>
 193 void process::pneigh_sort_by_load(const Compare& comp)
 194 {
 195     using std::placeholders::_1;
 196     using std::placeholders::_2;
 197     std::sort(pneigh.begin(), pneigh.end(),
 198               std::bind(comp,
 199                         std::bind(&neighbor::get_load, _1),
 200                         std::bind(&neighbor::get_load, _2)));
 201 }
 202
 203 #endif // !PROCESS_H
 204
 205 // Local variables:
 206 // mode: c++
 207 // End: