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 <tr1/functional>
   9 #ifdef USE_UNORDERED_MAP
  10 #  include <tr1/unordered_map>
  11 #  define MAP_TEMPLATE std::tr1::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     double get_data_send_amount() const    { return acc.data_send.amount; }
  38     double get_data_recv_amount() const    { return acc.data_recv.amount; }
  39     unsigned get_data_send_count() const   { return acc.data_send.count;  }
  40     unsigned get_data_recv_count() const   { return acc.data_recv.count;  }
  41     double get_ctrl_send_amount() const    { return acc.ctrl_send.amount; }
  42     double get_ctrl_recv_amount() const    { return acc.ctrl_recv.amount; }
  43     unsigned get_ctrl_send_count() const   { return acc.ctrl_send.count;  }
  44     unsigned get_ctrl_recv_count() const   { return acc.ctrl_recv.count;  }
  45
  46     int run();
  47
  48 protected:
  49     typedef std::vector<neighbor> neigh_type;
  50     typedef std::vector<neighbor*> pneigh_type;
  51
  52     pneigh_type pneigh;         // list of pointers to neighbors that
  53                                 // we are free to reorder
  54
  55     // Get and set current load, which may be real load, or expected
  56     // load if opt::bookkeeping is true.
  57     double get_load() const                { return expected_load; }
  58
  59     // The load balancing algorithm comes here...
  60     virtual void load_balance();
  61
  62     // Register some amount of load to send to given neighbor.
  63     void send(neighbor& nb, double amount);
  64     void send(neighbor* nb, double amount) { send(*nb, amount); }
  65
  66     // Sort pneigh by applying comp to their loads
  67     template <typename Compare>
  68     void pneigh_sort_by_load(const Compare& comp);
  69
  70     // Calls neighbor::print(verbose, logp, cat) for each member of neigh.
  71     void print_loads(bool verbose = false,
  72                      e_xbt_log_priority_t logp = xbt_log_priority_info,
  73                      xbt_log_category_t cat = _XBT_LOGV(default)) const;
  74
  75     // Calls neighbor::print(verbose, logp, cat) for each member of pneigh.
  76     void print_loads_p(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 private:
  81     static double total_load_init; // sum of process loads at init
  82     static double total_load_running; // sum of loads while running
  83     static double total_load_exit; // sum of process loads at exit
  84
  85     typedef MAP_TEMPLATE<m_host_t, neighbor*> rev_neigh_type;
  86     neigh_type neigh;           // list of neighbors (do not alter
  87                                 // after construction!)
  88     rev_neigh_type rev_neigh;   // map m_host_t -> neighbor
  89
  90     communicator comm;          // communicator for this process
  91     int ctrl_close_pending;     // number of "close" messages to wait
  92                                 // on ctrl channel
  93     int data_close_pending;     // number of "close" messages to wait
  94                                 // on data channel
  95     bool close_received;        // true if we received a "close" message
  96     bool finalizing;            // true when finalize() is running
  97
  98     unsigned lb_iter;           // counter of load-balancing iterations
  99     unsigned comp_iter;         // counter of computation iterations
 100
 101     double prev_load_broadcast; // used to ensure that we do not send
 102                                 // a same information messages
 103     double real_load;           // current load
 104     double expected_load;       // expected load in bookkeeping mode
 105
 106     mutex_t mutex;              // synchronization between threads
 107     condition_t cond;
 108
 109     struct mesg_accounting {
 110         double amount;          // sum of message size
 111         unsigned count;         // number of messages
 112         mesg_accounting(): amount(0.0), count(0) { }
 113     };
 114     struct accounting {
 115         double comp_amount;        // total computing done so far (flops)
 116         mesg_accounting data_send; // data messages sent
 117         mesg_accounting data_recv; // data messages received
 118         mesg_accounting ctrl_send; // ctrl message sent
 119         mesg_accounting ctrl_recv; // ctrl message received
 120         accounting(): comp_amount(0.0) { }
 121     };
 122     accounting acc;
 123
 124     void add_comp_amount(double amount) { acc.comp_amount += amount; }
 125     void add_data_send_mesg(double amount) {
 126         ++acc.data_send.count;
 127         acc.data_send.amount += amount;
 128     }
 129     void add_data_recv_mesg(double amount) {
 130         ++acc.data_recv.count;
 131         acc.data_recv.amount += amount;
 132     }
 133     void add_ctrl_send_mesg(double amount) {
 134         ++acc.ctrl_send.count;
 135         acc.ctrl_send.amount += amount;
 136     }
 137     void add_ctrl_recv_mesg(double amount) {
 138         ++acc.ctrl_recv.count;
 139         acc.ctrl_recv.amount += amount;
 140     }
 141
 142     // Load-balancing loop
 143     msg_thread* lb_thread;
 144     void load_balance_loop();
 145
 146     // Simulate computation loop
 147     void compute_loop();
 148
 149     // Check if we need to stop
 150     bool still_running();
 151
 152     // Returns the sum of "to_send" for all neighbors.
 153     double get_sum_of_to_send() const;
 154
 155     // Send procedures
 156     void ctrl_send(neighbor& nb);
 157     void data_send(neighbor& nb);
 158     void ctrl_close(neighbor& nb);
 159     void data_close(neighbor& nb);
 160
 161     // Receive procedure
 162     // Parameter "timeout" may be 0 for non-blocking operation, -1 for
 163     // infinite waiting, or any positive timeout.
 164     void ctrl_receive(double timeout);
 165     void data_receive(double timeout);
 166     void handle_message(message* msg, m_host_t from);
 167 };
 168
 169 template <typename Compare>
 170 void process::pneigh_sort_by_load(const Compare& comp)
 171 {
 172     using std::tr1::bind;
 173     using std::tr1::placeholders::_1;
 174     using std::tr1::placeholders::_2;
 175     std::sort(pneigh.begin(), pneigh.end(),
 176               bind(comp,
 177                    bind(&neighbor::get_load, _1),
 178                    bind(&neighbor::get_load, _2)));
 179 }
 180
 181 #endif // !PROCESS_H
 182
 183 // Local variables:
 184 // mode: c++
 185 // End: