Adding the torus network plugin with modelnet API, see...

Adding the torus network plugin with modelnet API, see model-net/README_MN_TEST.txt for instructions to run the test

codes/model-net.h

@@ -20,8 +20,8 @@ typedef struct torus_param torus_param;
 enum NETWORKS
 {
   SIMPLENET,
-  DRAGONFLY,
-  TORUS
+  TORUS,
+  DRAGONFLY
 };
 
 /* structs for initializing a network/ specifying network parameters */
@@ -42,9 +42,12 @@ struct dragonfly_param
 
 struct torus_param
 {
-  char* name;
-  int n_dim; /*Dimension of the torus network, 5-D, 7-D or any other*/
+  int n_dims; /*Dimension of the torus network, 5-D, 7-D or any other*/
   int* dim_length; /*Length of each torus dimension*/
+  double link_bandwidth;/* bandwidth for each torus link */
+  int buffer_size; /* number of buffer slots for each vc in flits*/
+  int num_vc; /* number of virtual channels for each torus link */
+  float mean_process;/* mean process time for each flit  */
 };
 /* NOTE: the following auxilliary functions are probably wrong; just leaving
  * these examples from simplenet for reference purposes.

codes/torus.h

+#ifndef INC_torus_h
+#define INC_torus_h
+
+#include <ross.h>
+#include <assert.h>
+
+#include "codes/codes_mapping.h"
+#include "codes/codes.h"
+#include "codes/model-net.h"
+#include "codes/model-net-method.h"
+
+#define CHUNK_SIZE 32
+#define DEBUG 1
+#define MEAN_INTERVAL 100
+#define CATEGORY_NAME_MAX 16
+#define MAX_NAME_LENGTH 256
+#define TRACE -1 
+
+// Total number of nodes in torus, calculate in main
+int N_nodes = 1;
+double link_bandwidth;
+int buffer_size;
+int num_vc;
+int n_dims;
+int * dim_length;
+int * factor;
+int * half_length;
+
+char grp_name[MAX_NAME_LENGTH], type_name[MAX_NAME_LENGTH];
+int grp_id, lp_type_id, rep_id, offset;
+
+typedef enum nodes_event_t nodes_event_t;
+typedef struct nodes_state nodes_state;
+typedef struct nodes_message nodes_message;
+
+/* Issues a torus packet event call */
+static void torus_packet_event(
+		       char* category,
+		       tw_lpid final_dest_lp,
+		       int packet_size,
+		       int remote_event_size,
+		       const void* remote_event,
+		       int self_event_size,
+		       const void* self_event,
+		       tw_lp *sender,
+		       int is_last_pckt);
+
+static void torus_packet_event_rc(tw_lp *sender);
+static void torus_setup(const void* net_params);
+static int torus_get_msg_sz(void);
+static const tw_lptype* torus_get_lp_type(void);
+
+/* data structure for torus statistics */
+struct model_net_method torus_method =
+{
+   .method_name = "torus",
+   .mn_setup = torus_setup,
+   .model_net_method_packet_event = torus_packet_event,
+   .model_net_method_packet_event_rc = torus_packet_event_rc,
+   .mn_get_lp_type = torus_get_lp_type,
+   .mn_get_msg_sz = torus_get_msg_sz,
+};
+
+enum nodes_event_t
+{
+  GENERATE = 1,
+  ARRIVAL, 
+  SEND,
+  CREDIT,
+};
+
+struct nodes_state
+{
+  unsigned long long packet_counter;            
+  tw_stime** next_link_available_time; 
+  tw_stime** next_credit_available_time;
+  tw_stime** next_flit_generate_time;
+  int** buffer;
+  int* dim_position;
+  int* neighbour_minus_lpID;
+  int* neighbour_plus_lpID;
+  int source_dim;
+  int direction;
+};
+
+struct nodes_message
+{
+  char category[CATEGORY_NAME_MAX];
+  tw_stime travel_start_time;
+  tw_stime saved_available_time;
+
+  unsigned long long packet_ID;
+  nodes_event_t	 type;
+
+  int saved_src_dim;
+  int saved_src_dir;
+
+  int* dest;
+
+  tw_lpid final_dest_gid;
+  tw_lpid dest_lp;
+  tw_lpid sender_lp;
+
+  int my_N_hop;
+  int source_dim;
+  int source_direction;
+  int next_stop;
+  int packet_size;
+  short chunk_id;
+
+  // for codes local and remote events
+  int local_event_size_bytes;
+  int remote_event_size_bytes;
+};
+
+tw_stime         average_travel_time = 0;
+tw_stime         total_time = 0;
+tw_stime         max_latency = 0;
+
+float head_delay=0.0;
+float credit_delay = 0.0;
+
+static unsigned long long       N_finished_packets = 0;
+static unsigned long long       total_hops = 0;
+// run time arguments
+int num_packets;
+int num_chunks;
+int packet_offset = 0;
+
+#endif

src/models/Makefile.subdir

 lib_LIBRARIES += src/libcodes-net.a
 
 nobase_include_HEADERS = \
- codes/model-net.h
+ codes/model-net.h	\
+ codes/torus.h		\
+ codes/model-net-method.h
 
 src_libcodes_net_a_SOURCES = \
  src/models/networks/model-net/model-net-method.h \
  src/models/networks/model-net/model-net.c \
- src/models/networks/model-net/simplenet-upd.c
+ src/models/networks/model-net/simplenet-upd.c \
+ src/models/networks/model-net/torus.c
 

src/models/networks/model-net/README_MN_TEST.txt

@@ -8,3 +8,5 @@ mpiexec -n 4 tests/modelnet-test --sync=2 tests/modelnet-test.conf
     <or>
 mpiexec -n 4 tests/modelnet-test --sync=3 tests/modelnet-test.conf
 
+(1)- To run the modelnet test with the simplenet network plugin, use tests/modelnet-test.conf (default setting).
+(2)- To run the modelnet test with the torus network plugin, use tests/modelnet-test-torus.conf file.

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

@@ -7,19 +7,20 @@
 #include <assert.h>
 
 #include "codes/model-net.h"
-#include "model-net-method.h"
+#include "codes/model-net-method.h"
 
 #define STR_SIZE 16
 #define PROC_TIME 10.0
 #define NUM_NETS 1
 
 extern struct model_net_method simplenet_method;
+extern struct model_net_method torus_method;
 //extern struct dragonfly_method dragonfly_method;
 //extern struct torus_method torus_method;
 
 /* Global array initialization, terminated with a NULL entry */
 static struct model_net_method* method_array[] =
-    {&simplenet_method, NULL};
+    {&simplenet_method, &torus_method, NULL};
 
 int model_net_setup(char* name,
 		    int packet_size,
@@ -63,7 +64,7 @@ void model_net_event(
      /*Determine the network name*/
      if(net_id < 0 || net_id > NUM_NETS)
      {
-        fprintf(stderr, "Error: undefined network ID %d (Available options 0 (simplenet), 1 (dragonfly) 2 (torus) ) \n", net_id);
+        fprintf(stderr, "Error: undefined network ID %d (Available options 0 (simplenet), 1 (torus) 2 (dragonfly) ) \n", net_id);
 	exit(-1);
      }
 
@@ -114,8 +115,8 @@ int model_net_set_params()
      
      configuration_get_value_double(&config, "PARAMS", "net_startup_ns", &net_startup_ns);
      configuration_get_value_double(&config, "PARAMS", "net_bw_mbps", &net_bw_mbps);
-     net_params.net_startup_ns = 1.5;
-     net_params.net_bw_mbps = 20000;
+     net_params.net_startup_ns = net_startup_ns;
+     net_params.net_bw_mbps =  net_bw_mbps;
      net_id = model_net_setup("simplenet", packet_size, (const void*)&net_params); /* Sets the network as simplenet and packet size 512 */
    }
   else if(strcmp("dragonfly", mn_name)==0)	  
@@ -124,7 +125,61 @@ int model_net_set_params()
     }
    else if(strcmp("torus", mn_name)==0)
      {
-	printf("\n not supported yet ");
+	torus_param net_params;
+	char dim_length[MAX_NAME_LENGTH];
+	int n_dims=0, buffer_size=0, num_vc=0, i=0;
+	double link_bandwidth=0;
+
+	configuration_get_value_int(&config, "PARAMS", "n_dims", &n_dims);
+	if(!n_dims)
+	{
+	   n_dims = 4; /* a 4-D torus */
+	   printf("\n Number of dimensions not specified, setting to %d ", n_dims);
+	}
+	
+	configuration_get_value_double(&config, "PARAMS", "link_bandwidth", &link_bandwidth);	
+	if(!link_bandwidth)
+	{
+		link_bandwidth = 2.0; /*default bg/q configuration */
+		printf("\n Link bandwidth not specified, setting to %lf ", link_bandwidth);
+	}
+
+	configuration_get_value_int(&config, "PARAMS", "buffer_size", &buffer_size);
+	if(!buffer_size)
+	{
+		buffer_size = 2048;
+		printf("\n Buffer size not specified, setting to %d ",buffer_size);
+	}
+
+	configuration_get_value_int(&config, "PARAMS", "num_vc", &num_vc);
+	if(!num_vc)
+	{
+		num_vc = 1; /*by default, we have one for taking packets, another for taking credit*/
+		printf("\n num_vc not specified, setting to %d ", num_vc);
+	}
+
+        configuration_get_value(&config, "PARAMS", "dim_length", dim_length, MAX_NAME_LENGTH);
+        char* token;
+	net_params.n_dims=n_dims;
+	net_params.num_vc=num_vc;
+	net_params.buffer_size=buffer_size;
+	net_params.link_bandwidth=link_bandwidth;
+	net_params.dim_length=malloc(n_dims*sizeof(int));
+        token = strtok(dim_length, ",");	
+	while(token != NULL)
+	{
+	   sscanf(token, "%d", &net_params.dim_length[i]);
+	   if(!net_params.dim_length[i])
+	   {
+	      printf("\n Invalid torus dimension specified %d, exitting... ", net_params.dim_length[i]);
+	      MPI_Finalize();
+	      exit(-1);
+	   }
+	   i++;
+
+	   token = strtok(NULL,",");
+	}
+	net_id = model_net_setup("torus", packet_size, (const void*)&net_params);
      }
   else
        printf("\n Invalid network argument %s ", mn_name);
@@ -203,6 +258,9 @@ void model_net_add_lp_type(int net_id)
        lp_type_register("modelnet_simplenet", model_net_get_lp_type(net_id));
    break;
 
+   case TORUS:
+       lp_type_register("modelnet_torus", model_net_get_lp_type(net_id));
+       break;
    default:
     {
         printf("\n Invalid net_id specified ");

src/models/networks/model-net/simplenet-upd.c

@@ -10,7 +10,7 @@
 
 #include "codes/lp-io.h"
 #include "codes/jenkins-hash.h"
-#include "model-net-method.h"
+#include "codes/model-net-method.h"
 #include "codes/model-net.h"
 #include "codes/codes_mapping.h"
 #include "codes/codes.h"

tests/modelnet-test-torus.conf

+LPGROUPS
+{
+   MODELNET_GRP
+   {
+      repetitions="16";
+      server="1";
+      modelnet_torus="1";
+   }
+}
+PARAMS
+{
+   packet_size="512";
+   modelnet="torus";
+   message_size="512";
+   n_dims="4";
+   dim_length="2,2,2,2";
+   link_bandwidth="2.0";
+   buffer_size="256";
+   num_vc="1";
+}

tests/modelnet-test.c

@@ -299,7 +299,7 @@ static void svr_finalize(
     svr_state * ns,
     tw_lp * lp)
 {
-    printf("server %llu recvd %d bytes in %f seconds, %f MiB/s sent_count %d recvd_count %d local_count %d \n", (unsigned long long)lp->gid/2, PAYLOAD_SZ*ns->msg_recvd_count, ns_to_s((tw_now(lp)-ns->start_ts)), 
+    printf("server %llu recvd %d bytes in %f seconds, %f MiB/s sent_count %d recvd_count %d local_count %d \n", (unsigned long long)lp->gid, PAYLOAD_SZ*ns->msg_recvd_count, ns_to_s((tw_now(lp)-ns->start_ts)), 
         ((double)(PAYLOAD_SZ*NUM_REQS)/(double)(1024*1024)/ns_to_s(tw_now(lp)-ns->start_ts)), ns->msg_sent_count, ns->msg_recvd_count, ns->local_recvd_count);
     return;
 }
@@ -463,8 +463,8 @@ static void handle_req_event(
     //printf("handle_req_event(), lp %llu src %llu .\n", (unsigned long long)lp->gid, (unsigned long long) m->src);
 
     /* safety check that this request got to the right server */
+//    printf("\n m->src %d lp->gid %d ", m->src, lp->gid);
     assert(lp->gid == (m->src + 2)%(NUM_SERVERS*2));
-
     ns->msg_recvd_count++;
 
     /* send ack back */

tests/modelnet-test.conf

@@ -9,8 +9,8 @@ LPGROUPS
 }
 PARAMS
 {
+   packet_size="512";
    modelnet="simplenet";
-   message_size="256";
    net_startup_ns="1.5";
    net_bw_mbps="20000";
 }