changes related to the ROSS model-level sampling feature

codes/lp-type-lookup.h

@@ -21,8 +21,8 @@ const tw_lptype* lp_type_lookup(const char* name);
 /* register an LP with CODES/ROSS */
 void lp_type_register(const char* name, const tw_lptype* type);
 
-void trace_type_register(const char* name, const st_trace_type* type);
-const st_trace_type* trace_type_lookup(const char* name);
+void st_model_type_register(const char* name, const st_model_types* type);
+const st_model_types* st_model_type_lookup(const char* name);
 #ifdef __cplusplus
 }
 #endif

codes/model-net-method.h

@@ -63,8 +63,8 @@ struct model_net_method
     revent_f mn_sample_rc_fn;
     init_f mn_sample_init_fn;
     final_f mn_sample_fini_fn;
-    void (*mn_trace_register)(st_trace_type *base_type);
-    const st_trace_type* (*mn_get_trace_type)();
+    void (*mn_model_stat_register)(st_model_types *base_type);
+    const st_model_types* (*mn_get_model_stat_types)();
 };
 
 extern struct model_net_method * method_array[];

codes/model-net.h

@@ -388,7 +388,7 @@ void model_net_set_msg_param(
 
 /* returns pointer to LP information for simplenet module */
 const tw_lptype* model_net_get_lp_type(int net_id);
-const st_trace_type* model_net_get_trace_type(int net_id);
+const st_model_types* model_net_get_model_stat_type(int net_id);
 
 DEPRECATED
 uint64_t model_net_get_packet_size(int net_id);

codes/net/dragonfly.h

@@ -91,6 +91,8 @@ struct terminal_message
    tw_stime saved_hist_start_time;
    tw_stime saved_sample_time;
    tw_stime msg_start_time;
+   tw_stime saved_busy_time_ross;
+   tw_stime saved_fin_chunks_ross;
 
    int saved_hist_num;
    int saved_occupancy;

src/network-workloads/model-net-mpi-replay.c

@@ -1569,22 +1569,29 @@ void nw_lp_event_collect(nw_message *m, tw_lp *lp, char *buffer, int *collect_fl
     memcpy(buffer, &type, sizeof(type));
 }
 
-st_trace_type nw_lp_trace_types[] = {
+void nw_lp_model_stat_collect(nw_state *s, tw_lp *lp, char *buffer)
+{
+    return;
+}
+
+st_model_types nw_lp_model_types[] = {
     {(rbev_trace_f) nw_lp_event_collect,
      sizeof(int),
      (ev_trace_f) nw_lp_event_collect,
-     sizeof(int)},
+     sizeof(int),
+     (model_stat_f) nw_lp_model_stat_collect,
+     0},
     {0}
 };
 
-static const st_trace_type  *nw_lp_get_trace_types(void)
+static const st_model_types  *nw_lp_get_model_stat_types(void)
 {
-    return(&nw_lp_trace_types[0]);
+    return(&nw_lp_model_types[0]);
 }
 
-void nw_lp_register_trace()
+void nw_lp_register_model()
 {
-    trace_type_register("nw-lp", nw_lp_get_trace_types());
+    st_model_type_register("nw-lp", nw_lp_get_model_stat_types());
 }
 /* end of ROSS event tracing setup */
 
@@ -1659,8 +1666,8 @@ int main( int argc, char** argv )
    nw_add_lp_type();
    model_net_register();
 
-    if (g_st_ev_trace)
-        nw_lp_register_trace();
+    if (g_st_ev_trace || g_st_model_stats)
+        nw_lp_register_model();
 
    net_ids = model_net_configure(&num_nets);
 //   assert(num_nets == 1);

src/network-workloads/model-net-synthetic-fattree.c

@@ -130,22 +130,29 @@ void ft_svr_event_collect(svr_msg *m, tw_lp *lp, char *buffer, int *collect_flag
     memcpy(buffer, &type, sizeof(type));
 }
 
-st_trace_type ft_svr_trace_types[] = {
+void ft_svr_model_stat_collect(svr_state *s, tw_lp *lp, char *buffer)
+{
+    return;
+}
+
+st_model_types ft_svr_model_types[] = {
     {(rbev_trace_f) ft_svr_event_collect,
      sizeof(int),
      (ev_trace_f) ft_svr_event_collect,
-     sizeof(int)},
+     sizeof(int),
+     (model_stat_f) ft_svr_model_stat_collect,
+     0},
     {0}
 };
 
-static const st_trace_type  *ft_svr_get_trace_types(void)
+static const st_model_types  *ft_svr_get_model_stat_types(void)
 {
-    return(&ft_svr_trace_types[0]);
+    return(&ft_svr_model_types[0]);
 }
 
-void ft_svr_register_trace()
+void ft_svr_register_model_stats()
 {
-    trace_type_register("server", ft_svr_get_trace_types());
+    st_model_type_register("server", ft_svr_get_model_stat_types());
 }
 
 const tw_optdef app_opt [] =
@@ -437,7 +444,7 @@ int main(
     svr_add_lp_type();
 
     if (g_st_ev_trace)
-        ft_svr_register_trace();
+        ft_svr_register_model_stats();
 
     codes_mapping_setup();
 

src/network-workloads/model-net-synthetic.c

@@ -116,22 +116,29 @@ void svr_event_collect(svr_msg *m, tw_lp *lp, char *buffer, int *collect_flag)
     memcpy(buffer, &type, sizeof(type));
 }
 
-st_trace_type svr_trace_types[] = {
+void svr_model_stat_collect(svr_state *s, tw_lp *lp, char *buffer)
+{
+    return;
+}
+
+st_model_types svr_model_types[] = {
     {(rbev_trace_f) svr_event_collect,
      sizeof(int),
      (ev_trace_f) svr_event_collect,
-     sizeof(int)},
+     sizeof(int),
+     (model_stat_f) svr_model_stat_collect,
+     0},
     {0}
 };
 
-static const st_trace_type  *svr_get_trace_types(void)
+static const st_model_types  *svr_get_model_stat_types(void)
 {
-    return(&svr_trace_types[0]);
+    return(&svr_model_types[0]);
 }
 
-void svr_register_trace()
+void svr_register_model_types()
 {
-    trace_type_register("server", svr_get_trace_types());
+    st_model_type_register("server", svr_get_model_stat_types());
 }
 
 const tw_optdef app_opt [] =
@@ -393,8 +400,8 @@ int main(
     model_net_register();
     svr_add_lp_type();
 
-    if (g_st_ev_trace)
-        svr_register_trace();
+    if (g_st_ev_trace || g_st_model_stats)
+        svr_register_model_types();
 
     codes_mapping_setup();
 

src/networks/model-net/fattree.c

@@ -1811,6 +1811,7 @@ void switch_packet_receive_rc(switch_state * s,
 	s_arrive_r++;
 #endif
     int output_port = msg->saved_vc;
+    tw_rand_reverse_unif(lp->rng);
     if(bf->c1)
     {
         tw_rand_reverse_unif(lp->rng);
@@ -2636,10 +2637,13 @@ int ft_get_output_port( switch_state * s, tw_bf * bf, fattree_message * msg,
 
   assert(end_port > start_port);
 
-  outport = start_port;
+  //outport = start_port;
+  // when occupancy is same, just choose random port
+  outport = tw_rand_integer(lp->rng, start_port, end_port-1);  
   int load = s->vc_occupancy[outport] + s->queued_length[outport];
   if(load != 0) {
-    for(int port = start_port + 1; port < end_port; port++) {
+    //for(int port = start_port + 1; port < end_port; port++) {
+    for(int port = start_port; port < end_port; port++) {
       if(s->vc_occupancy[port] +  s->queued_length[port] < load) {
         load = s->vc_occupancy[port] +  s->queued_length[port];
         outport = port;
@@ -2968,23 +2972,27 @@ void fattree_event_collect(fattree_message *m, tw_lp *lp, char *buffer, int *col
     memcpy(buffer, &type, sizeof(type));
 }
 
-st_trace_type fattree_trace_types[] = {
+// TODO will need to separate fattree_method into one for terminal and one for switch
+// in order to use the ROSS model stats collection
+st_model_types fattree_model_types[] = {
     {(rbev_trace_f) fattree_event_collect,
      sizeof(int),
      (ev_trace_f) fattree_event_collect,
-     sizeof(int)},
+     sizeof(int),
+    NULL,
+    0},
     {0}
 };
 
-static const st_trace_type  *fattree_get_trace_types(void)
+static const st_model_types  *fattree_get_model_stat_types(void)
 {
-    return(&fattree_trace_types[0]);
+    return(&fattree_model_types[0]);
 }
 
-static void fattree_register_trace(st_trace_type *base_type)
+static void fattree_register_model_stats(st_model_types *base_type)
 {
-    trace_type_register(LP_CONFIG_NM, base_type);
-    trace_type_register("fattree_switch", &fattree_trace_types[0]);
+    st_model_type_register(LP_CONFIG_NM, base_type);
+    st_model_type_register("fattree_switch", &fattree_model_types[0]);
     //trace_type_register("fattree_switch", base_type);
 }
 /*** END of ROSS event tracing additions */
@@ -3018,7 +3026,7 @@ struct model_net_method fattree_method =
 //  .model_net_method_find_local_device = NULL,
   .mn_collective_call = NULL,
   .mn_collective_call_rc = NULL,
-  .mn_trace_register = fattree_register_trace,
-  .mn_get_trace_type = fattree_get_trace_types
+  .mn_model_stat_register = fattree_register_model_stats,
+  .mn_get_model_stat_types = fattree_get_model_stat_types
 };
 

src/networks/model-net/model-net-lp.c

@@ -53,7 +53,7 @@ typedef struct model_net_base_state {
     // lp type and state of underlying model net method - cache here so we
     // don't have to constantly look up
     const tw_lptype *sub_type;
-    const st_trace_type *sub_trace_type;
+    const st_model_types *sub_model_type;
     void *sub_state;
 } model_net_base_state;
 
@@ -139,10 +139,10 @@ void mn_event_collect(model_net_wrap_msg *m, tw_lp *lp, char *buffer, int *colle
             break;
         case MN_BASE_PASS:
             sub_msg = ((char*)m)+msg_offsets[((model_net_base_state*)lp->cur_state)->net_id];
-            if (g_st_ev_trace == RB_TRACE)
-                (((model_net_base_state*)lp->cur_state)->sub_trace_type->rbev_trace)(sub_msg, lp, buffer, collect_flag);
+            if (g_st_ev_trace == RB_TRACE || g_st_ev_trace == COMMIT_TRACE)
+                (((model_net_base_state*)lp->cur_state)->sub_model_type->rbev_trace)(sub_msg, lp, buffer, collect_flag);
             else if (g_st_ev_trace == FULL_TRACE)
-                (((model_net_base_state*)lp->cur_state)->sub_trace_type->ev_trace)(sub_msg, lp, buffer, collect_flag);
+                (((model_net_base_state*)lp->cur_state)->sub_model_type->ev_trace)(sub_msg, lp, buffer, collect_flag);
             break;
         default:  // this shouldn't happen, but can help detect an issue
             type = 9004;
@@ -150,11 +150,22 @@ void mn_event_collect(model_net_wrap_msg *m, tw_lp *lp, char *buffer, int *colle
     }
 }
 
-st_trace_type mn_trace_types = {
+void mn_model_stat_collect(model_net_base_state *s, tw_lp *lp, char *buffer)
+{
+    // need to call the model level stats collection fn
+    (*s->sub_model_type->model_stat_fn)(s->sub_state, lp, buffer);
+    return;
+}
+
+st_model_types mn_model_types[MAX_NETS];
+
+st_model_types mn_model_base_type = {
     (rbev_trace_f) mn_event_collect,
      sizeof(int),
      (ev_trace_f) mn_event_collect,
      sizeof(int),
+     (model_stat_f) mn_model_stat_collect,
+     0
 };
 
 /**** END LP, EVENT PROCESSING FUNCTION DECLS ****/
@@ -195,12 +206,14 @@ void model_net_base_register(int *do_config_nets){
                         &model_net_base_lp);
             else
                 method_array[i]->mn_register(&model_net_base_lp);
-            if (g_st_ev_trace) // for ROSS event tracing
+            if (g_st_ev_trace || g_st_model_stats) // for ROSS event tracing
             {
-                if (method_array[i]->mn_trace_register == NULL)
-                    trace_type_register(model_net_lp_config_names[i], &mn_trace_types);
+                memcpy(&mn_model_types[i], &mn_model_base_type, sizeof(st_model_types));
+
+                if (method_array[i]->mn_model_stat_register == NULL)
+                    st_model_type_register(model_net_lp_config_names[i], &mn_model_types[i]);
                 else
-                    method_array[i]->mn_trace_register(&mn_trace_types);
+                    method_array[i]->mn_model_stat_register(&mn_model_types[i]);
             }
         }
     }
@@ -404,8 +417,12 @@ void model_net_base_lp_init(
 
     ns->sub_type = model_net_get_lp_type(ns->net_id);
 
-    if (g_st_ev_trace)
-        ns->sub_trace_type = model_net_get_trace_type(ns->net_id);
+    /* some ROSS instrumentation setup */
+    if (g_st_ev_trace || g_st_model_stats)
+    {
+        ns->sub_model_type = model_net_get_model_stat_type(ns->net_id);
+        mn_model_types[ns->net_id].mstat_sz = ns->sub_model_type->mstat_sz;
+    }
 
     // NOTE: some models actually expect LP state to be 0 initialized...
     // *cough anything that uses mn_stats_array cough*

src/networks/model-net/model-net.c

@@ -597,9 +597,9 @@ const tw_lptype* model_net_get_lp_type(int net_id)
    return method_array[net_id]->mn_get_lp_type();
 }
 
-const st_trace_type* model_net_get_trace_type(int net_id)
+const st_model_types* model_net_get_model_stat_type(int net_id)
 {
-   return method_array[net_id]->mn_get_trace_type();
+   return method_array[net_id]->mn_get_model_stat_types();
 }
 
 void model_net_report_stats(int net_id)

src/util/codes_mapping.c

@@ -458,7 +458,7 @@ static void codes_mapping_init(void)
      tw_lpid nkp_per_pe = g_tw_nkp;
      tw_lpid         lpid, kpid;
      const tw_lptype *lptype;
-     const st_trace_type *trace_type;
+     const st_model_types *trace_type;
 
      /* have 16 kps per pe, this is the optimized configuration for ROSS custom mapping */
      for(kpid = 0; kpid < nkp_per_pe; kpid++)
@@ -489,10 +489,10 @@ static void codes_mapping_init(void)
          else
              /* sorry, const... */
              tw_lp_settype(ross_lid, (tw_lptype*) lptype);
-         if (g_st_ev_trace)
+         if (g_st_ev_trace || g_st_model_stats)
          {
-             trace_type = trace_type_lookup(lp_type_name);
-             st_evtrace_settype(ross_lid, (st_trace_type*) trace_type);
+             trace_type = st_model_type_lookup(lp_type_name);
+             st_model_settype(ross_lid, (st_model_types*) trace_type);
          }
      }
      return;

src/util/lp-type-lookup.c

@@ -16,7 +16,7 @@ struct lp_name_mapping
 {
     const char* name;
     const tw_lptype* type;
-    const st_trace_type* trace_type;
+    const st_model_types* trace_type;
 };
 
 static struct lp_name_mapping map_array[MAX_LP_TYPES];
@@ -48,7 +48,7 @@ const tw_lptype* lp_type_lookup(const char* name)
 }
 
 // needs to be called after lp_type_register()
-void trace_type_register(const char* name, const st_trace_type* type)
+void st_model_type_register(const char* name, const st_model_types* type)
 {
     int i;
 
@@ -61,7 +61,7 @@ void trace_type_register(const char* name, const st_trace_type* type)
     }
 }
 
-const st_trace_type* trace_type_lookup(const char* name)
+const st_model_types* st_model_type_lookup(const char* name)
 {
     int i;