model-net.c 12.1 KB
Newer Older
1
/*
Philip Carns's avatar
Philip Carns committed
2
 * Copyright (C) 2013 University of Chicago.
3
 * See COPYRIGHT notice in top-level directory.
Philip Carns's avatar
Philip Carns committed
4
 *
5
 */
Philip Carns's avatar
Philip Carns committed
6

7
8
9
#include <string.h>
#include <assert.h>

10
#include "codes/model-net.h"
11
#include "codes/model-net-method.h"
12
13
14

#define STR_SIZE 16
#define PROC_TIME 10.0
15
#define NUM_NETS 2
16
17

extern struct model_net_method simplenet_method;
18
extern struct model_net_method torus_method;
19
extern struct model_net_method dragonfly_method;
20
21
22

/* Global array initialization, terminated with a NULL entry */
static struct model_net_method* method_array[] =
23
    {&simplenet_method, &torus_method, &dragonfly_method, NULL};
24

25
static int model_net_get_msg_sz(int net_id);
26

27
28
29
30
31
32
33
34
35
36
37
38
int model_net_setup(char* name,
		    int packet_size,
		    const void* net_params)
{
     int i;
    /* find struct for underlying method (according to configuration file) */
     for(i=0; method_array[i] != NULL; i++)
     {
     	if(strcmp(method_array[i]->method_name, name) == 0)
	{
	   method_array[i]->mn_setup(net_params);
	   method_array[i]->packet_size = packet_size;
39
	   model_net_add_lp_type(i);
40
41
42
	   return(i);
	}
     }
43
     fprintf(stderr, "Error: undefined network name %s (Available options simplenet, torus, dragonfly) \n", name);
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
     return -1; // indicating error
}

void model_net_event(
    int net_id,
    char* category, 
    tw_lpid final_dest_lp, 
    int message_size, 
    int remote_event_size,
    const void* remote_event,
    int self_event_size,
    const void* self_event,
    tw_lp *sender)
{
    /* determine packet size for underlying method */
     int packet_size = model_net_get_packet_size(net_id);
     int num_packets = message_size/packet_size; /* Number of packets to be issued by the API */
     int i;
     int last = 0;

Philip Carns's avatar
Philip Carns committed
64
65
66
67
68
69
70
71
72
     if((remote_event_size + self_event_size + model_net_get_msg_sz(net_id))
        > g_tw_msg_sz)
     {
        fprintf(stderr, "Error: model_net trying to transmit an event of size %d but ROSS is configured for events of size %zd\n",
            (remote_event_size + self_event_size + model_net_get_msg_sz(net_id)),
            g_tw_msg_sz);
        abort();
     }

73
74
75
76
77
78
     if(message_size % packet_size)
	num_packets++; /* Handle the left out data if message size is not exactly divisible by packet size */

     /*Determine the network name*/
     if(net_id < 0 || net_id > NUM_NETS)
     {
79
        fprintf(stderr, "Error: undefined network ID %d (Available options 0 (simplenet), 1 (torus) 2 (dragonfly) ) \n", net_id);
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
	exit(-1);
     }

    /* issue N packets using method API */
    /* somehow mark the final packet as the one responsible for delivering
     * the self event and remote event 
     *
     * local event is delivered to caller of this function, remote event is
     * passed along through network hops and delivered to final_dest_lp
     */

     for( i = 0; i < num_packets; i++ )
       {
	  /*Mark the last packet to the net method API*/
	   if(i == num_packets - 1)
           {
	      last = 1;
              /* also calculate the last packet's size */
              packet_size = message_size - ((num_packets-1)*packet_size);
            }
	  /* Number of packets and packet ID is passed to the underlying network to mark the final packet for local event completion*/
	  method_array[net_id]->model_net_method_packet_event(category, final_dest_lp, packet_size, remote_event_size, remote_event, self_event_size, self_event, sender, last);	  
       }
    return;
}

106
107
108
109
int model_net_set_params()
{
  char mn_name[MAX_NAME_LENGTH];
  int packet_size = 0;
110
  int net_id=-1;
111
112
113
114
115

  config_lpgroups_t paramconf;
  configuration_get_lpgroups(&config, "PARAMS", &paramconf);
  configuration_get_value(&config, "PARAMS", "modelnet", mn_name, MAX_NAME_LENGTH);
  configuration_get_value_int(&config, "PARAMS", "packet_size", &packet_size);
116

117
118
119
120
121
122
123
124
125
126
127
128
  if(!packet_size)
  {
	packet_size = 512;
	printf("\n Warning, no packet size specified, setting packet size to %d ", packet_size);
  }
  if(strcmp("simplenet",mn_name)==0)
   {
     double net_startup_ns, net_bw_mbps;
     simplenet_param net_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);
129
130
     net_params.net_startup_ns = net_startup_ns;
     net_params.net_bw_mbps =  net_bw_mbps;
131
132
133
134
     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)	  
    {
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
       dragonfly_param net_params;
       int num_routers=0, num_vcs=0, local_vc_size=0, global_vc_size=0, cn_vc_size=0;
       double local_bandwidth=0.0, cn_bandwidth=0.0, global_bandwidth=0.0;
       
       configuration_get_value_int(&config, "PARAMS", "num_routers", &num_routers);
       if(!num_routers)
	{
	   num_routers = 4; 
	   printf("\n Number of dimensions not specified, setting to %d ", num_routers);
        } 
       net_params.num_routers = num_routers; 

       configuration_get_value_int(&config, "PARAMS", "num_vcs", &num_vcs);
       if(!num_vcs)
       {
          num_vcs = 1;
	  printf("\n Number of virtual channels not specified, setting to %d ", num_vcs);
       }
       net_params.num_vcs = num_vcs;

       configuration_get_value_int(&config, "PARAMS", "local_vc_size", &local_vc_size);
       if(!local_vc_size)
	{
	   local_vc_size = 1024;
	   printf("\n Buffer size of local channels not specified, setting to %d ", local_vc_size);
	}
       net_params.local_vc_size = local_vc_size;

       configuration_get_value_int(&config, "PARAMS", "global_vc_size", &global_vc_size);
       if(!global_vc_size)
	{
	  global_vc_size = 2048;
	  printf("\n Buffer size of global channels not specified, setting to %d ", global_vc_size);
	}
       net_params.global_vc_size = global_vc_size;

       configuration_get_value_int(&config, "PARAMS", "cn_vc_size", &cn_vc_size);
       if(!cn_vc_size)
	 {
	    cn_vc_size = 1024;
	    printf("\n Buffer size of compute node channels not specified, setting to %d ", cn_vc_size);
	 }
       net_params.cn_vc_size = cn_vc_size;

	configuration_get_value_double(&config, "PARAMS", "local_bandwidth", &local_bandwidth);
        if(!local_bandwidth)
	  {
	    local_bandwidth = 5.25;
	    printf("\n Bandwidth of local channels not specified, setting to %lf ", local_bandwidth);
	 }
       net_params.local_bandwidth = local_bandwidth;

       configuration_get_value_double(&config, "PARAMS", "global_bandwidth", &global_bandwidth);
        if(!global_bandwidth)
	{
	     global_bandwidth = 4.7;
	     printf("\n Bandwidth of global channels not specified, setting to %lf ", global_bandwidth);
	}
	net_params.global_bandwidth = global_bandwidth;

	configuration_get_value_double(&config, "PARAMS", "cn_bandwidth", &cn_bandwidth);
	if(!cn_bandwidth)
	 {
	     cn_bandwidth = 5.25;
	     printf("\n Bandwidth of compute node channels not specified, setting to %lf ", cn_bandwidth);
	}
	net_params.cn_bandwidth = cn_bandwidth;

       char routing[MAX_NAME_LENGTH];
       configuration_get_value(&config, "PARAMS", "routing", routing, MAX_NAME_LENGTH);
       if(strcmp(routing, "minimal") == 0)
	   net_params.routing = 0;
       else if(strcmp(routing, "nonminimal")==0 || strcmp(routing,"non-minimal")==0)
	       net_params.routing = 1;
       else
       {
       	   printf("\n No routing protocol specified, setting to minimal routing");
   	   net_params.routing = 0;	   
       }
    net_id = model_net_setup("dragonfly", packet_size, (const void*)&net_params);   
215
216
217
    }
   else if(strcmp("torus", mn_name)==0)
     {
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
	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);
272
273
274
275
276
     }
  else
       printf("\n Invalid network argument %s ", mn_name);
  return net_id;
}
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
void model_net_event_rc(
    int net_id,
    tw_lp *sender,
    int message_size)
{
    /* this will be used for reverse computation of anything calculated
     * within th model_net_event() function call itself (not reverse
     * handling for the underlying methods, which will have their own events
     * and reverse handlers
     */
    int packet_size = model_net_get_packet_size(net_id);
    int num_packets = message_size/packet_size; /* For rolling back */
    int i;

    if(message_size % packet_size)
      num_packets++;

     for( i = 0; i < num_packets; i++ )
       {
	  /* Number of packets and packet ID is passed to the underlying network to mark the final packet for local event completion*/
	  method_array[net_id]->model_net_method_packet_event_rc(sender);	  
       }
    return;
} 

/* returns the message size, can be either simplenet, dragonfly or torus message size*/
303
static int model_net_get_msg_sz(int net_id)
304
305
306
307
308
309
310
311
312
313
314
315
316
{
   // TODO: Add checks on network name
   // TODO: Add dragonfly and torus network models
   if(net_id < 0 || net_id > NUM_NETS)
     {
      printf("%s Error: Uninitializied modelnet network, call modelnet_init first\n", __FUNCTION__);
      exit(-1);
     }

       return method_array[net_id]->mn_get_msg_sz();
}

/* returns the packet size in the modelnet struct */
317
int model_net_get_packet_size(int net_id)
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
{
  if(net_id < 0 || net_id > NUM_NETS)
     {
       fprintf(stderr, "%s Error: Uninitializied modelnet network, call modelnet_init first\n", __FUNCTION__);
       exit(-1);
     }
  return method_array[net_id]->packet_size; // TODO: where to set the packet size?
}

/* returns lp type for modelnet */
const tw_lptype* model_net_get_lp_type(int net_id)
{
    if(net_id < 0 || net_id > NUM_NETS)
     {
       fprintf(stderr, "%s Error: Uninitializied modelnet network, call modelnet_init first\n", __FUNCTION__);
       exit(-1);
     }

   // TODO: ADd checks by network names
   // Add dragonfly and torus network models
   return method_array[net_id]->mn_get_lp_type();
}

341
342
343
344
345
346
347
348
349
350
351
352
void model_net_report_stats(int net_id)
{
  if(net_id < 0 || net_id > NUM_NETS)
  {
    fprintf(stderr, "%s Error: Uninitializied modelnet network, call modelnet_init first\n", __FUNCTION__);
    exit(-1);
   }

     // TODO: ADd checks by network names
     //    // Add dragonfly and torus network models
   return method_array[net_id]->mn_report_stats();
}
353
354
355
356
357
358
359
360
361
/* registers the lp type */
void model_net_add_lp_type(int net_id)
{
 switch(net_id)
 {
   case SIMPLENET:
       lp_type_register("modelnet_simplenet", model_net_get_lp_type(net_id));
   break;

362
363
364
   case TORUS:
       lp_type_register("modelnet_torus", model_net_get_lp_type(net_id));
       break;
365
366
367
   case DRAGONFLY:
       lp_type_register("modelnet_dragonfly", model_net_get_lp_type(net_id));
       break;
368
   default:
369
370
371
372
373
374
375
376
377
378
379
380
381
382
    {
        printf("\n Invalid net_id specified ");
	exit(-1);
    }
 }
}
/*
 * Local variables:
 *  c-indent-level: 4
 *  c-basic-offset: 4
 * End:
 *
 * vim: ft=c ts=8 sts=4 sw=4 expandtab
 */