Adding multiple background traffic patterns

scripts/gen-cray-topo/theta.py

@@ -30,8 +30,8 @@ intercon = open(sys.argv[3], "wb")
 def router(group, row, col):
     return group*96 + row*16 + col
 
-numblack = np.zeros((864,864), dtype=np.int)
-numblue = np.zeros((864,864), dtype=np.int)
+numblack = np.zeros((960,960), dtype=np.int)
+numblue = np.zeros((960,960), dtype=np.int)
 
 with open(filename) as ofile:
     matches =    re.findall('c\d+-\dc\ds\d+a0l\d+\((\d+):(\d):(\d+)\).(\w+).->.c\d+-\dc\ds\d+a0l\d+\((\d+):(\d):(\d+)\)', ofile.read(), re.MULTILINE)

src/network-workloads/conf/dragonfly-custom/modelnet-test-dragonfly-1728-nodes.conf.in

@@ -2,7 +2,7 @@ LPGROUPS
 {
    MODELNET_GRP
    {
-      repetitions="864";
+      repetitions="960";
 # name of this lp changes according to the model
       nw-lp="2";
 # these lp names will be the same for dragonfly-custom model
@@ -27,7 +27,7 @@ PARAMS
    # number of router columns 
    num_router_cols="16";
    # number of groups in the network
-   num_groups="9";
+   num_groups="10";
 # buffer size in bytes for local virtual channels 
    local_vc_size="8192";
 #buffer size in bytes for global virtual channels 

src/network-workloads/conf/dragonfly-custom/modelnet-test-dragonfly-theta.conf

@@ -2,7 +2,7 @@ LPGROUPS
 {
    MODELNET_GRP
    {
-      repetitions="864";
+      repetitions="960";
 # name of this lp changes according to the model
       nw-lp="4";
 # these lp names will be the same for dragonfly-custom model
@@ -27,7 +27,7 @@ PARAMS
    # number of router columns 
    num_router_cols="16";
    # number of groups in the network
-   num_groups="9";
+   num_groups="10";
 # buffer size in bytes for local virtual channels 
    local_vc_size="8192";
 #buffer size in bytes for global virtual channels 
@@ -48,9 +48,9 @@ PARAMS
 # number of global channels per router 
    num_global_channels="4";
 # network config file for intra-group connections 
-   intra-group-connections="../src/network-workloads/conf/dragonfly-custom/intra-theta";
+   intra-group-connections="../src/network-workloads/conf/dragonfly-custom/intra-theta-upd";
 # network config file for inter-group connections
-   inter-group-connections="../src/network-workloads/conf/dragonfly-custom/inter-theta";
+   inter-group-connections="../src/network-workloads/conf/dragonfly-custom/inter-theta-upd";
 # routing protocol to be used 
    routing="adaptive";
 }

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

@@ -28,7 +28,6 @@
 #define lprintf(_fmt, ...) \
         do {if (CS_LP_DBG) printf(_fmt, __VA_ARGS__);} while (0)
 #define MAX_STATS 65536
-#define PAYLOAD_SZ 1024
 
 static int msg_size_hash_compare(
             void *key, struct qhash_head *link);
@@ -55,6 +54,7 @@ static int64_t EAGER_THRESHOLD = 8192;
 static int alloc_spec = 0;
 static tw_stime self_overhead = 10.0;
 static tw_stime mean_interval = 100000;
+static int payload_sz = 1024;
 
 /* Doing LP IO*/
 static char lp_io_dir[256] = {'\0'};
@@ -138,6 +138,14 @@ enum MPI_NW_EVENTS
     CLI_NBR_FINISH,
 };
 
+/* type of synthetic traffic */
+enum TRAFFIC
+{
+    UNIFORM = 1, /* sends message to a randomly selected node */
+    NEAREST_NEIGHBOR = 2, /* sends message to the next node (potentially connected to the same router) */
+    ALLTOALL = 3, /* sends message to all other nodes */
+    STENCIL = 4  /* sends message to 4 nearby neighbors */
+};
 struct mpi_workload_sample
 {
     /* Sampling data */
@@ -203,6 +211,7 @@ struct nw_state
     int app_id;
     int local_rank;
 
+    int synthetic_pattern;
     int is_finished;
     int neighbor_completed;
 
@@ -309,6 +318,7 @@ struct nw_message
        double saved_wait_time;
        double saved_delay;
        int64_t saved_num_bytes;
+       int saved_syn_length;
    } rc;
 };
 
@@ -591,12 +601,13 @@ static void gen_synthetic_tr_rc(nw_state * s, tw_bf * bf, nw_message * m, tw_lp
     if(bf->c0)
         return;
 
-    model_net_event_rc2(lp, &m->event_rc);
-    s->gen_data -= PAYLOAD_SZ;
-
-    num_syn_bytes_sent -= PAYLOAD_SZ;
-    tw_rand_reverse_unif(lp->rng);
-    tw_rand_reverse_unif(lp->rng);
+    int i;
+    for (i=0; i < m->rc.saved_syn_length; i++){
+        model_net_event_rc2(lp, &m->event_rc);
+        s->gen_data -= payload_sz;
+        num_syn_bytes_sent -= payload_sz;
+    }
+        tw_rand_reverse_unif(lp->rng);
 
 }
 
@@ -611,38 +622,105 @@ static void gen_synthetic_tr(nw_state * s, tw_bf * bf, nw_message * m, tw_lp * l
 
     /* Get job information */
     tw_lpid global_dest_id;
+    int intm_dest_id;
+    nw_message remote_m;
 
     struct codes_jobmap_id jid;
     jid = codes_jobmap_to_local_id(s->nw_id, jobmap_ctx); 
 
     int num_clients = codes_jobmap_get_num_ranks(jid.job, jobmap_ctx);
-    int dest_svr = tw_rand_integer(lp->rng, 0, num_clients - 1);
 
-    if(dest_svr == s->local_rank)
+    /* Find destination */
+    int* dest_svr = NULL; 
+    int i, length=0;
+    switch(s->synthetic_pattern)
     {
-       dest_svr = (s->local_rank + 1) % num_clients;
-    }
-   
-    jid.rank = dest_svr;
-
-    int intm_dest_id = codes_jobmap_to_global_id(jid, jobmap_ctx); 
-    global_dest_id = codes_mapping_get_lpid_from_relative(intm_dest_id, NULL, NW_LP_NM, NULL, 0);
-
-    nw_message remote_m;
-    remote_m.fwd.sim_start_time = tw_now(lp);
-    remote_m.fwd.dest_rank = dest_svr;
-    remote_m.msg_type = CLI_BCKGND_ARRIVE;
-    remote_m.fwd.num_bytes = PAYLOAD_SZ;
-    remote_m.fwd.app_id = s->app_id;
-    remote_m.fwd.src_rank = s->local_rank;
-
-    m->event_rc = model_net_event(net_id, "synthetic-tr", global_dest_id, PAYLOAD_SZ, 0.0, 
-            sizeof(nw_message), (const void*)&remote_m, 
-            0, NULL, lp);
-    
-    s->gen_data += PAYLOAD_SZ;
-    num_syn_bytes_sent += PAYLOAD_SZ; 
+        case UNIFORM:
+        {
+            length = 1;
+            dest_svr = (int*) calloc(1, sizeof(int));
+            dest_svr[0] = tw_rand_integer(lp->rng, 0, num_clients - 1);
+            if(dest_svr[0] == s->local_rank)
+                dest_svr[0] = (s->local_rank + 1) % num_clients;
+        }
+        break;
+        case NEAREST_NEIGHBOR:
+        {
+            length = 1;
+            dest_svr = (int*) calloc(1, sizeof(int));
+            dest_svr[0] = (s->local_rank + 1) % num_clients;
+        }
+        break;
+        case ALLTOALL:
+        {
+            dest_svr = (int*) calloc(num_clients-1, sizeof(int));
+            int index = 0;
+            for (i=0;i<num_clients;i++)
+            {
+                if(i!=s->local_rank) 
+                {
+                    dest_svr[index] = i;
+                    index++;
+                    length++;
+                }
+            }
+        }
+        break;
+        case STENCIL:  //2D 4-point stencil
+        {
+            /* I think this code snippet is coming from the LLNL stencil patterns. */
+            int digits, x=1, y=1, row, col, temp=num_clients;
+            length = 4;
+            dest_svr = (int*) calloc(4, sizeof(int));
+            for (digits = 0; temp > 0; temp >>= 1)
+                digits++;
+            digits = digits/2;
+            for (i = 0; i < digits; i++)
+                x = x * 2;
+            y = num_clients / x;
+            //printf("\nStencil Syn: x=%d, y=%d", x, y);
+            row = s->local_rank / y;
+            col = s->local_rank % y;
+
+            dest_svr[0] = row * y + ((col-1+y)%y);   /* left neighbor */
+            dest_svr[1] = row * y + ((col+1+y)%y);   /* right neighbor */
+            dest_svr[2] = ((row-1+x)%x) * y + col;   /* bottom neighbor */
+            dest_svr[3] = ((row+1+x)%x) * y + col;   /* up neighbor */
+        }
+        break;
+        default:
+            tw_error(TW_LOC, "Undefined traffic pattern");
+    }   
+    /* Record length for reverse handler*/
+    m->rc.saved_syn_length = length;
 
+    if(length > 0)
+    {
+        // m->event_array_rc = (model_net_event_return) malloc(length * sizeof(model_net_event_return));
+        //printf("\nRANK %d Dests %d", s->local_rank, length);
+        for (i = 0; i < length; i++)
+        {
+            /* Generate synthetic traffic */
+            jid.rank = dest_svr[i];
+            intm_dest_id = codes_jobmap_to_global_id(jid, jobmap_ctx); 
+            global_dest_id = codes_mapping_get_lpid_from_relative(intm_dest_id, NULL, NW_LP_NM, NULL, 0);
+
+            remote_m.fwd.sim_start_time = tw_now(lp);
+            remote_m.fwd.dest_rank = dest_svr[i];
+            remote_m.msg_type = CLI_BCKGND_ARRIVE;
+            remote_m.fwd.num_bytes = payload_sz;
+            remote_m.fwd.app_id = s->app_id;
+            remote_m.fwd.src_rank = s->local_rank;
+
+            // printf("\nAPP %d SRC %d Dest %d (twid %llu)", jid.job, s->local_rank, dest_svr[i], global_dest_id);
+            m->event_rc = model_net_event(net_id, "synthetic-tr", global_dest_id, payload_sz, 0.0, 
+                    sizeof(nw_message), (const void*)&remote_m, 
+                    0, NULL, lp);
+            
+            s->gen_data += payload_sz;
+            num_syn_bytes_sent += payload_sz; 
+        }
+    }
     /* New event after MEAN_INTERVAL */  
     tw_stime ts = mean_interval  + tw_rand_exponential(lp->rng, noise); 
     tw_event * e;
@@ -651,6 +729,8 @@ static void gen_synthetic_tr(nw_state * s, tw_bf * bf, nw_message * m, tw_lp * l
     m_new = (struct nw_message*)tw_event_data(e);
     m_new->msg_type = CLI_BCKGND_GEN;
     tw_event_send(e);
+
+    free(dest_svr);
 }
 
 void arrive_syn_tr_rc(nw_state * s, tw_bf * bf, nw_message * m, tw_lp * lp)
@@ -1845,6 +1925,18 @@ void nw_test_init(nw_state* s, tw_lp* lp)
 
    if(strcmp(file_name_of_job[lid.job], "synthetic") == 0)
    {
+        int synthetic_pattern;
+        sscanf(file_name_of_job[lid.job], "synthetic%d", &synthetic_pattern);
+        if(synthetic_pattern <=0 || synthetic_pattern > 4)
+        {
+            printf("\n Undefined synthetic pattern: setting to uniform random ");
+            s->synthetic_pattern = 1;
+        }
+        else
+        {
+            s->synthetic_pattern = synthetic_pattern;
+        }
+
         tw_event * e;
         nw_message * m_new;
         tw_stime ts = tw_rand_exponential(lp->rng, mean_interval/1000);