process.h

   1 #ifndef PROCESS_H
   2 #define PROCESS_H
   3
   4 #include <algorithm>
   5 #include <atomic>
   6 #include <unordered_map>
   7 #include <vector>
   8 #include <simgrid/msg.h>
   9 #include <xbt/log.h>
  10 #include "communicator.h"
  11 #include "misc.h"
  12 #include "msg_thread.h"
  13 #include "neighbor.h"
  14 #include "options.h"
  15 #include "simgrid_features.h"
  16 #include "synchro.h"
  17
  18 class process {
  19 public:
  20     static void set_proc_mutex(mutex_t* m) { proc_mutex = m;            }
  21
  22     // Note: normally used with proc_mutex locked.
  23     static double get_total_load_init()    { return total_load_init;    }
  24     static double get_total_load_running() { return total_load_running; }
  25     static double get_total_load_exit()    { return total_load_exit;    }
  26
  27     process(int argc, char* argv[]);
  28     virtual ~process();
  29
  30     double get_real_load() const           { return real_load;            }
  31     double get_comp_amount() const         { return acc.comp_amount;      }
  32     unsigned get_comp_iter() const         { return comp_iter;            }
  33     unsigned get_all_comp_iter() const     { return all_comp_iter;        }
  34     double get_iter_deviation() const;
  35     double get_data_send_amount() const    { return acc.data_send.amount; }
  36     double get_data_recv_amount() const    { return acc.data_recv.amount; }
  37     unsigned get_data_send_count() const   { return acc.data_send.count;  }
  38     unsigned get_data_recv_count() const   { return acc.data_recv.count;  }
  39     double get_ctrl_send_amount() const    { return acc.ctrl_send.amount; }
  40     double get_ctrl_recv_amount() const    { return acc.ctrl_recv.amount; }
  41     unsigned get_ctrl_send_count() const   { return acc.ctrl_send.count;  }
  42     unsigned get_ctrl_recv_count() const   { return acc.ctrl_recv.count;  }
  43     double get_idle_duration() const       { return idle_duration;        }
  44     double get_convergence() const         { return convergence;          }
  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 mutex_t *proc_mutex; // protect access to global variables
  82                                 // (must be set before constructing
  83                                 // the first object!)
  84
  85     static double total_load_init; // sum of process loads at init
  86     static double total_load_running; // sum of loads while running
  87     static double total_load_exit; // sum of process loads at exit
  88
  89     static int process_counter;
  90     static double total_load_average;
  91     static double average_load_ratio;
  92     static double load_diff_threshold;
  93
  94     static std::atomic<int> convergence_counter;
  95
  96     typedef std::unordered_map<msg_host_t, neighbor*> rev_neigh_type;
  97     neigh_type neigh;           // list of neighbors (do not alter
  98                                 // after construction!)
  99     rev_neigh_type rev_neigh;   // map msg_host_t -> neighbor
 100
 101     communicator comm;          // communicator for this process
 102     int ctrl_close_pending;     // number of "close" messages to wait
 103                                 // on ctrl channel
 104     int data_close_pending;     // number of "close" messages to wait
 105                                 // on data channel
 106     bool close_received;        // true if we received a "close" message
 107     bool finalizing;            // true when finalize() is running
 108
 109     unsigned lb_iter;           // counter of load-balancing iterations
 110     unsigned comp_iter;         // counter of computation iterations
 111     unsigned all_comp_iter;     // counter of computation iterations
 112                                 // (counting empty iterations too)
 113
 114     double prev_load_broadcast; // used to ensure that we do not send
 115                                 // a same information messages
 116     double real_load;           // current load
 117     double expected_load;       // expected load in bookkeeping mode
 118     double received_load;       // load received from neighbors
 119
 120     double idle_duration;       // how long we had nothing to compute
 121     double convergence;         // date when convergence was achieved, or -1.0
 122     int local_convergence_counter; // number of iterations since convergence
 123
 124     mutex_t mutex;              // synchronization between threads
 125     condition_t cond;
 126
 127     struct mesg_accounting {
 128         double amount;          // sum of message size
 129         unsigned count;         // number of messages
 130         mesg_accounting(): amount(0.0), count(0) { }
 131     };
 132     struct accounting {
 133         double comp_amount;        // total computing done so far (flops)
 134         mesg_accounting data_send; // data messages sent
 135         mesg_accounting data_recv; // data messages received
 136         mesg_accounting ctrl_send; // ctrl message sent
 137         mesg_accounting ctrl_recv; // ctrl message received
 138         accounting(): comp_amount(0.0) { }
 139     };
 140     accounting acc;             // use a structure so that it is
 141                                 // automatically initialized a
 142                                 // construction
 143
 144     void add_comp_amount(double amount) { acc.comp_amount += amount; }
 145     void add_data_send_mesg(double amount) {
 146         ++acc.data_send.count;
 147         acc.data_send.amount += amount;
 148     }
 149     void add_data_recv_mesg(double amount) {
 150         ++acc.data_recv.count;
 151         acc.data_recv.amount += amount;
 152     }
 153     void add_ctrl_send_mesg(double amount) {
 154         ++acc.ctrl_send.count;
 155         acc.ctrl_send.amount += amount;
 156     }
 157     void add_ctrl_recv_mesg(double amount) {
 158         ++acc.ctrl_recv.count;
 159         acc.ctrl_recv.amount += amount;
 160     }
 161
 162     // Load-balancing loop
 163     msg_thread* lb_thread;
 164     void load_balance_loop();
 165
 166     // Simulate computation loop
 167     void compute_loop();
 168
 169     // Check for convergence
 170     void convergence_check();
 171
 172     // Check if we need to stop
 173     bool still_running();
 174
 175     // Returns the sum of "to_send" for all neighbors.
 176     double get_sum_of_to_send() const;
 177
 178     // Compute load_to_send (for data_send), subject to the execution parameters
 179     static double compute_load_to_send(double desired);
 180
 181     // Send procedures
 182     void ctrl_send(neighbor& nb);
 183     void data_send(neighbor& nb);
 184     void ctrl_close(neighbor& nb);
 185     void data_close(neighbor& nb);
 186
 187     // Receive procedure
 188     // Parameter "timeout" may be 0 for non-blocking operation, -1 for
 189     // infinite waiting, or any positive timeout.
 190     void ctrl_receive(double timeout);
 191     void data_receive(double timeout);
 192     void handle_message(message* msg, msg_host_t from);
 193 };
 194
 195 template <typename Compare>
 196 void process::pneigh_sort_by_load(const Compare& comp)
 197 {
 198     std::sort(pneigh.begin(), pneigh.end(),
 199               [&comp](const neighbor* a, const neighbor* b) {
 200                   return comp(a->get_load(), b->get_load());
 201               });
 202 }
 203
 204 #endif // !PROCESS_H
 205
 206 // Local variables:
 207 // mode: c++
 208 // End: