model-net.c 17.8 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 PULL_MSG_SIZE 128

15 16 17 18
#define STR_SIZE 16
#define PROC_TIME 10.0

extern struct model_net_method simplenet_method;
19
extern struct model_net_method simplewan_method;
20
extern struct model_net_method torus_method;
21
extern struct model_net_method dragonfly_method;
Philip Carns's avatar
Philip Carns committed
22
extern struct model_net_method loggp_method;
23 24 25

/* Global array initialization, terminated with a NULL entry */
static struct model_net_method* method_array[] =
26
    {&simplenet_method, &simplewan_method, &torus_method, &dragonfly_method, &loggp_method, NULL};
27

28 29 30
int in_sequence = 0;
tw_stime mn_msg_offset = 0.0;

31
static int model_net_get_msg_sz(int net_id);
32

33
int model_net_setup(char* name,
34
		    uint64_t packet_size,
35 36
		    const void* net_params)
{
37
     int i;
38 39 40 41 42 43 44
    /* 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;
45
	   model_net_add_lp_type(i);
46 47 48
	   return(i);
	}
     }
49
     fprintf(stderr, "Error: undefined network name %s (Available options simplenet, torus, dragonfly) \n", name);
50 51 52
     return -1; // indicating error
}

53 54 55 56 57 58 59 60 61 62
int model_net_get_id(char *name){
    int i;
    for(i=0; method_array[i] != NULL; i++) {
        if(strcmp(method_array[i]->method_name, name) == 0) {
            return i;
        }
    }
    return -1;
}

63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144
void model_net_write_stats(tw_lpid lpid, struct mn_stats* stat)
{
    int ret;
    char id[32];
    char data[1024];

    sprintf(id, "model-net-category-%s", stat->category);
    sprintf(data, "lp:%ld\tsend_count:%ld\tsend_bytes:%ld\tsend_time:%f\t" 
        "recv_count:%ld\trecv_bytes:%ld\trecv_time:%f\tmax_event_size:%ld\n",
        (long)lpid,
        stat->send_count,
        stat->send_bytes,
        stat->send_time,
        stat->recv_count,
        stat->recv_bytes,
        stat->recv_time,
        stat->max_event_size);

    ret = lp_io_write(lpid, id, strlen(data), data);
    assert(ret == 0);

    return;
}

void model_net_print_stats(tw_lpid lpid, mn_stats mn_stats_array[])
{

    int i;
    struct mn_stats all;

    memset(&all, 0, sizeof(all));
    sprintf(all.category, "all");

    for(i=0; i<CATEGORY_MAX; i++)
    {
        if(strlen(mn_stats_array[i].category) > 0)
        {
            all.send_count += mn_stats_array[i].send_count;
            all.send_bytes += mn_stats_array[i].send_bytes;
            all.send_time += mn_stats_array[i].send_time;
            all.recv_count += mn_stats_array[i].recv_count;
            all.recv_bytes += mn_stats_array[i].recv_bytes;
            all.recv_time += mn_stats_array[i].recv_time;
            if(mn_stats_array[i].max_event_size > all.max_event_size)
                all.max_event_size = mn_stats_array[i].max_event_size;

            model_net_write_stats(lpid, &mn_stats_array[i]);
        }
    }
    model_net_write_stats(lpid, &all);
}

struct mn_stats* model_net_find_stats(const char* category, mn_stats mn_stats_array[])
{
    int i;
    int new_flag = 0;
    int found_flag = 0;

    for(i=0; i<CATEGORY_MAX; i++)
    {
        if(strlen(mn_stats_array[i].category) == 0)
        {
            found_flag = 1;
            new_flag = 1;
            break;
        }
        if(strcmp(category, mn_stats_array[i].category) == 0)
        {
            found_flag = 1;
            new_flag = 0;
            break;
        }
    }
    assert(found_flag);

    if(new_flag)
    {
        strcpy(mn_stats_array[i].category, category);
    }
    return(&mn_stats_array[i]);
}

145 146 147 148 149 150 151 152 153 154 155 156 157
static void model_net_event_impl(
        int net_id,
        char* category, 
        tw_lpid final_dest_lp, 
        uint64_t message_size, 
        int is_pull,
        uint64_t pull_msg_size,
        tw_stime offset,
        int remote_event_size,
        const void* remote_event,
        int self_event_size,
        const void* self_event,
        tw_lp *sender) {
158
    /* determine packet size for underlying method */
159 160 161
     uint64_t packet_size = model_net_get_packet_size(net_id);
     uint64_t num_packets = message_size/packet_size; /* Number of packets to be issued by the API */
     uint64_t i;
162 163
     int last = 0;

164
     //printf("\n number of packets %d message size %d ", num_packets, message_size);
165 166 167 168 169 170 171 172 173
     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();
     }

174 175 176
     if(message_size % packet_size)
	num_packets++; /* Handle the left out data if message size is not exactly divisible by packet size */

177 178 179
     if(message_size < packet_size)
         num_packets = 1;

180
     /*Determine the network name*/
181
     if(net_id < 0 || net_id >= MAX_NETS)
182
     {
183
        fprintf(stderr, "Error: undefined network ID %d (Available options 0 (simplenet), 1 (torus) 2 (dragonfly) ) \n", net_id);
184 185 186 187 188 189 190 191 192 193 194
	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
     */

195
     tw_stime poffset = (in_sequence) ? mn_msg_offset : 0.0;
196 197 198 199 200 201 202 203 204 205
     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*/
206 207 208
	  poffset += method_array[net_id]->model_net_method_packet_event(category,
                  final_dest_lp, packet_size, is_pull, pull_msg_size, poffset+offset,
                  remote_event_size, remote_event, self_event_size, self_event, sender, last);
209
       }
210
    if (in_sequence) mn_msg_offset = poffset;
211 212 213
    return;
}

214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248

void model_net_event(
    int net_id,
    char* category, 
    tw_lpid final_dest_lp, 
    uint64_t message_size, 
    tw_stime offset,
    int remote_event_size,
    const void* remote_event,
    int self_event_size,
    const void* self_event,
    tw_lp *sender)
{
    model_net_event_impl(net_id, category, final_dest_lp, message_size, 0, 0,
            offset, remote_event_size, remote_event, self_event_size,
            self_event, sender); 
}

void model_net_pull_event(
        int net_id,
        char *category,
        tw_lpid final_dest_lp,
        uint64_t message_size,
        tw_stime offset,
        int self_event_size,
        const void *self_event,
        tw_lp *sender){
    /* NOTE: for a pull, we are filling the *remote* event - it will be remote
     * from the destination's POV */
    model_net_event_impl(net_id, category, final_dest_lp, PULL_MSG_SIZE, 1,
            message_size, offset, self_event_size, self_event, 0, NULL,
            sender); 
}


249 250 251
int model_net_set_params()
{
  char mn_name[MAX_NAME_LENGTH];
252 253
  long int packet_size_l = 0;
  uint64_t packet_size;
254
  int net_id=-1;
255 256 257 258

  config_lpgroups_t paramconf;
  configuration_get_lpgroups(&config, "PARAMS", &paramconf);
  configuration_get_value(&config, "PARAMS", "modelnet", mn_name, MAX_NAME_LENGTH);
259 260
  configuration_get_value_longint(&config, "PARAMS", "packet_size", &packet_size_l);
  packet_size = packet_size_l;
261

262 263 264
  if(!packet_size)
  {
	packet_size = 512;
265
	printf("\n Warning, no packet size specified, setting packet size to %llu ", packet_size);
266 267 268 269 270 271 272 273
  }
  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);
274 275
     net_params.net_startup_ns = net_startup_ns;
     net_params.net_bw_mbps =  net_bw_mbps;
276 277
     net_id = model_net_setup("simplenet", packet_size, (const void*)&net_params); /* Sets the network as simplenet and packet size 512 */
   }
278 279 280 281 282 283
  else if (strcmp("simplewan",mn_name)==0){
    simplewan_param net_params;
    configuration_get_value_relpath(&config, "PARAMS", "net_startup_ns_file", net_params.startup_filename, MAX_NAME_LENGTH);
    configuration_get_value_relpath(&config, "PARAMS", "net_bw_mbps_file", net_params.bw_filename, MAX_NAME_LENGTH);
    net_id = model_net_setup("simplewan", packet_size, (const void*)&net_params);
  }
Philip Carns's avatar
Philip Carns committed
284 285 286 287 288
   else if(strcmp("loggp",mn_name)==0)
   {
     char net_config_file[256];
     loggp_param net_params;
     
289
     configuration_get_value_relpath(&config, "PARAMS", "net_config_file", net_config_file, 256);
Philip Carns's avatar
Philip Carns committed
290 291 292 293
     net_params.net_config_file = net_config_file;
     net_id = model_net_setup("loggp", packet_size, (const void*)&net_params); /* Sets the network as loggp and packet size 512 */
   }

294 295
  else if(strcmp("dragonfly", mn_name)==0)	  
    {
296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375
       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);   
376 377 378
    }
   else if(strcmp("torus", mn_name)==0)
     {
379 380
	torus_param net_params;
	char dim_length[MAX_NAME_LENGTH];
381
	int n_dims=0, buffer_size=0, num_vc=0, i=0, chunk_size = 0;
382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404
	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);
	}

405 406 407 408 409 410
	configuration_get_value_int(&config, "PARAMS", "chunk_size", &chunk_size);
	if(!chunk_size)
	{
	       chunk_size = 32;
	       printf("\n Chunk size not specified, setting to %d ", chunk_size);
	}
411 412 413 414 415 416 417 418 419 420 421 422
	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;
423
	net_params.chunk_size = chunk_size;
424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439
	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);
440 441 442 443 444
     }
  else
       printf("\n Invalid network argument %s ", mn_name);
  return net_id;
}
445
static void model_net_event_impl_rc(
446 447
    int net_id,
    tw_lp *sender,
448
    uint64_t message_size)
449 450 451 452 453 454
{
    /* 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
     */
455
    uint64_t packet_size = model_net_get_packet_size(net_id);
456 457 458 459 460 461 462 463 464 465 466 467 468 469
    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;
} 

470 471 472 473 474 475 476 477 478 479 480 481 482
void model_net_event_rc(
        int net_id,
        tw_lp *sender,
        uint64_t message_size){
    model_net_event_impl_rc(net_id,sender,message_size);
}

void model_net_pull_event_rc(
        int net_id,
        tw_lp *sender) {
    model_net_event_impl_rc(net_id, sender, PULL_MSG_SIZE);
}

483
/* returns the message size, can be either simplenet, dragonfly or torus message size*/
484
static int model_net_get_msg_sz(int net_id)
485 486 487
{
   // TODO: Add checks on network name
   // TODO: Add dragonfly and torus network models
488
   if(net_id < 0 || net_id >= MAX_NETS)
489 490 491 492 493 494 495 496 497
     {
      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 */
498
uint64_t model_net_get_packet_size(int net_id)
499
{
500
  if(net_id < 0 || net_id >= MAX_NETS)
501 502 503 504 505 506 507 508 509 510
     {
       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)
{
511
    if(net_id < 0 || net_id >= MAX_NETS)
512 513 514 515 516 517 518 519 520 521
     {
       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();
}

522 523
void model_net_report_stats(int net_id)
{
524
  if(net_id < 0 || net_id >= MAX_NETS)
525 526 527 528 529 530 531
  {
    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
532 533
   method_array[net_id]->mn_report_stats();
   return;
534
}
535 536 537 538 539 540 541 542
/* 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;
543 544 545
   case SIMPLEWAN:
        lp_type_register("modelnet_simplewan", model_net_get_lp_type(net_id));
        break;
546 547 548
   case TORUS:
       lp_type_register("modelnet_torus", model_net_get_lp_type(net_id));
       break;
549 550 551
   case DRAGONFLY:
       lp_type_register("modelnet_dragonfly", model_net_get_lp_type(net_id));
       break;
552 553 554
   case LOGGP:
       lp_type_register("modelnet_loggp", model_net_get_lp_type(net_id));
       break;
555
   default:
556 557 558 559 560 561
    {
        printf("\n Invalid net_id specified ");
	exit(-1);
    }
 }
}
562 563 564 565 566 567

tw_lpid model_net_find_local_device(int net_id, tw_lp *sender)
{
    return(method_array[net_id]->model_net_method_find_local_device(sender));
}

568 569 570 571 572 573 574 575
/*
 * Local variables:
 *  c-indent-level: 4
 *  c-basic-offset: 4
 * End:
 *
 * vim: ft=c ts=8 sts=4 sw=4 expandtab
 */