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 void set_proc_mutex(mutex_t* m) { proc_mutex = m;            }
  29
  30     // Note: normally used with proc_mutex locked.
  31     static double get_total_load_init()    { return total_load_init;    }
  32     static double get_total_load_running() { return total_load_running; }
  33     static double get_total_load_exit()    { return total_load_exit;    }
  34
  35     process(int argc, char* argv[]);
  36     virtual ~process();
  37
  38     double get_real_load() const           { return real_load;            }
  39     double get_comp_amount() const         { return acc.comp_amount;      }
  40     unsigned get_comp_iter() const         { return comp_iter;            }
  41     unsigned get_all_comp_iter() const     { return all_comp_iter;        }
  42     double get_iter_deviation() const;
  43     double get_data_send_amount() const    { return acc.data_send.amount; }
  44     double get_data_recv_amount() const    { return acc.data_recv.amount; }
  45     unsigned get_data_send_count() const   { return acc.data_send.count;  }
  46     unsigned get_data_recv_count() const   { return acc.data_recv.count;  }
  47     double get_ctrl_send_amount() const    { return acc.ctrl_send.amount; }
  48     double get_ctrl_recv_amount() const    { return acc.ctrl_recv.amount; }
  49     unsigned get_ctrl_send_count() const   { return acc.ctrl_send.count;  }
  50     unsigned get_ctrl_recv_count() const   { return acc.ctrl_recv.count;  }
  51     double get_idle_duration() const       { return idle_duration;        }
  52     double get_convergence() const         { return convergence;          }
  53
  54     int run();
  55
  56 protected:
  57     typedef std::vector<neighbor> neigh_type;
  58     typedef std::vector<neighbor*> pneigh_type;
  59
  60     pneigh_type pneigh;         // list of pointers to neighbors that
  61                                 // we are free to reorder
  62
  63     // Get and set current load, which may be real load, or expected
  64     // load if opt::bookkeeping is true.
  65     double get_load() const                { return expected_load; }
  66
  67     // The load balancing algorithm comes here...
  68     virtual void load_balance();
  69
  70     // Register some amount of load to send to given neighbor.
  71     void send(neighbor& nb, double amount);
  72     void send(neighbor* nb, double amount) { send(*nb, amount); }
  73
  74     // Sort pneigh by applying comp to their loads
  75     template <typename Compare>
  76     void pneigh_sort_by_load(const Compare& comp);
  77
  78     // Calls neighbor::print(verbose, logp, cat) for each member of neigh.
  79     void print_loads(bool verbose = false,
  80                      e_xbt_log_priority_t logp = xbt_log_priority_info,
  81                      xbt_log_category_t cat = _XBT_LOGV(default)) const;
  82
  83     // Calls neighbor::print(verbose, logp, cat) for each member of pneigh.
  84     void print_loads_p(bool verbose = false,
  85                        e_xbt_log_priority_t logp = xbt_log_priority_info,
  86                        xbt_log_category_t cat = _XBT_LOGV(default)) const;
  87
  88 private:
  89     static mutex_t *proc_mutex; // protect access to global variables
  90                                 // (must be set before constructing
  91                                 // the first object!)
  92
  93     static double total_load_init; // sum of process loads at init
  94     static double total_load_running; // sum of loads while running
  95     static double total_load_exit; // sum of process loads at exit
  96
  97     static int process_counter;
  98     static double total_load_average;
  99     static double load_diff_threshold;
 100
 101     typedef MAP_TEMPLATE<m_host_t, neighbor*> rev_neigh_type;
 102     neigh_type neigh;           // list of neighbors (do not alter
 103                                 // after construction!)
 104     rev_neigh_type rev_neigh;   // map m_host_t -> neighbor
 105
 106     communicator comm;          // communicator for this process
 107     int ctrl_close_pending;     // number of "close" messages to wait
 108                                 // on ctrl channel
 109     int data_close_pending;     // number of "close" messages to wait
 110                                 // on data channel
 111     bool close_received;        // true if we received a "close" message
 112     bool finalizing;            // true when finalize() is running
 113
 114     unsigned lb_iter;           // counter of load-balancing iterations
 115     unsigned comp_iter;         // counter of computation iterations
 116     unsigned all_comp_iter;     // counter of computation iterations
 117                                 // (counting empty iterations too)
 118
 119     double prev_load_broadcast; // used to ensure that we do not send
 120                                 // a same information messages
 121     double real_load;           // current load
 122     double expected_load;       // expected load in bookkeeping mode
 123     double received_load;       // load received from neighbors
 124
 125     double idle_duration;       // how long we had nothing to compute
 126     double convergence;         // date when convergence was achieved, or -1.0
 127
 128     mutex_t mutex;              // synchronization between threads
 129     condition_t cond;
 130
 131     struct mesg_accounting {
 132         double amount;          // sum of message size
 133         unsigned count;         // number of messages
 134         mesg_accounting(): amount(0.0), count(0) { }
 135     };
 136     struct accounting {
 137         double comp_amount;        // total computing done so far (flops)
 138         mesg_accounting data_send; // data messages sent
 139         mesg_accounting data_recv; // data messages received
 140         mesg_accounting ctrl_send; // ctrl message sent
 141         mesg_accounting ctrl_recv; // ctrl message received
 142         accounting(): comp_amount(0.0) { }
 143     };
 144     accounting acc;             // use a structure so that it is
 145                                 // automatically initialized a
 146                                 // construction
 147
 148     void add_comp_amount(double amount) { acc.comp_amount += amount; }
 149     void add_data_send_mesg(double amount) {
 150         ++acc.data_send.count;
 151         acc.data_send.amount += amount;
 152     }
 153     void add_data_recv_mesg(double amount) {
 154         ++acc.data_recv.count;
 155         acc.data_recv.amount += amount;
 156     }
 157     void add_ctrl_send_mesg(double amount) {
 158         ++acc.ctrl_send.count;
 159         acc.ctrl_send.amount += amount;
 160     }
 161     void add_ctrl_recv_mesg(double amount) {
 162         ++acc.ctrl_recv.count;
 163         acc.ctrl_recv.amount += amount;
 164     }
 165
 166     // Load-balancing loop
 167     msg_thread* lb_thread;
 168     void load_balance_loop();
 169
 170     // Simulate computation loop
 171     void compute_loop();
 172
 173     // Check for convergence
 174     void convergence_check();
 175
 176     // Check if we need to stop
 177     bool still_running();
 178
 179     // Returns the sum of "to_send" for all neighbors.
 180     double get_sum_of_to_send() const;
 181
 182     // Compute load_to_send (for data_send), subject to the execution parameters
 183     static double compute_load_to_send(double desired);
 184
 185     // Send procedures
 186     void ctrl_send(neighbor& nb);
 187     void data_send(neighbor& nb);
 188     void ctrl_close(neighbor& nb);
 189     void data_close(neighbor& nb);
 190
 191     // Receive procedure
 192     // Parameter "timeout" may be 0 for non-blocking operation, -1 for
 193     // infinite waiting, or any positive timeout.
 194     void ctrl_receive(double timeout);
 195     void data_receive(double timeout);
 196     void handle_message(message* msg, m_host_t from);
 197 };
 198
 199 template <typename Compare>
 200 void process::pneigh_sort_by_load(const Compare& comp)
 201 {
 202     using std::placeholders::_1;
 203     using std::placeholders::_2;
 204     std::sort(pneigh.begin(), pneigh.end(),
 205               std::bind(comp,
 206                         std::bind(&neighbor::get_load, _1),
 207                         std::bind(&neighbor::get_load, _2)));
 208 }
 209
 210 #endif // !PROCESS_H
 211
 212 // Local variables:
 213 // mode: c++
 214 // End: