simplewan.c 29.6 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14
/*
 * Copyright (C) 2014 University of Chicago.
 * See COPYRIGHT notice in top-level directory.
 *
 */

#include <string.h>
#include <assert.h>
#include <ross.h>

#include "codes/lp-io.h"
#include "codes/jenkins-hash.h"
#include "codes/model-net-method.h"
#include "codes/model-net.h"
15
#include "codes/model-net-lp.h"
16 17
#include "codes/codes_mapping.h"
#include "codes/codes.h"
18
#include "codes/net/simplewan.h"
19 20 21 22

#define CATEGORY_NAME_MAX 16
#define CATEGORY_MAX 12

23 24
#define SIMPLEWAN_DEBUG 0

25 26 27
#define LP_CONFIG_NM (model_net_lp_config_names[SIMPLEWAN])
#define LP_METHOD_NM (model_net_method_names[SIMPLEWAN])

28 29 30 31 32 33 34 35 36 37
// parameters for simplewan configuration
struct simplewan_param
{
    double * net_startup_ns_table;
    double * net_bw_mbps_table;
    int mat_len;
    int num_lps;
};
typedef struct simplewan_param simplewan_param;

38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58
/*Define simplewan data types and structs*/
typedef struct sw_state sw_state;

typedef struct category_idles_s category_idles;

struct category_idles_s{
    /* each simplewan "NIC" actually has N connections, so we need to track
     * idle times across all of them to correctly do stats */
    tw_stime send_next_idle_all;
    tw_stime send_prev_idle_all;
    tw_stime recv_next_idle_all;
    tw_stime recv_prev_idle_all;
    char category[CATEGORY_NAME_MAX];
};

struct sw_state
{
    /* next idle times for network card, both inbound and outbound */
    tw_stime *send_next_idle;
    tw_stime *recv_next_idle;

59 60 61
    const char * anno;
    const simplewan_param * params;

62 63 64 65 66 67 68 69 70 71 72
    int id; /* logical id for matrix lookups */

    /* Each simplewan "NIC" actually has N connections, so we need to track
     * idle times across all of them to correctly do stats.
     * Additionally need to track different idle times across different 
     * categories */
    category_idles idle_times_cat[CATEGORY_MAX];

    struct mn_stats sw_stats_array[CATEGORY_MAX];
};

73 74 75 76 77
/* annotation-specific parameters (unannotated entry occurs at the 
 * last index) */
static uint64_t                  num_params = 0;
static simplewan_param         * all_params = NULL;
static const config_anno_map_t * anno_map   = NULL;
78 79 80 81 82 83 84 85 86 87 88

static int sw_magic = 0;

/* returns a pointer to the lptype struct to use for simplewan LPs */
static const tw_lptype* sw_get_lp_type(void);

/* set model parameters:
 * - startup_fname - path containing triangular matrix of net latencies, in ns
 * - bw_fname      - path containing triangular matrix of bandwidths in MB/s.
 * note that this merely stores the files, they will be parsed later 
 */
89 90 91 92 93 94
static void sw_set_params(
        const char      * startup_fname,
        const char      * bw_fname,
        simplewan_param * params);

static void sw_configure();
95 96 97 98 99 100 101 102 103 104 105 106

/* retrieve the size of the portion of the event struct that is consumed by
 * the simplewan module.  The caller should add this value to the size of
 * its own event structure to get the maximum total size of a message.
 */
static int sw_get_msg_sz(void);

/* Returns the simplewan magic number */
static int sw_get_magic();

/* given two simplewan logical ids, do matrix lookups to get the point-to-point
 * latency/bandwidth */
107 108 109 110 111
static double sw_get_table_ent(
        int      from_id, 
        int      to_id,
        int      num_lps,
        double * table);
112 113 114 115 116

/* category lookup */
static category_idles* sw_get_category_idles(
        char * category, category_idles *idles);

117 118 119 120 121 122
/* collective network calls */
static void simple_wan_collective();

/* collective network calls-- rc */
static void simple_wan_collective_rc();

123 124
/* Issues a simplewan packet event call */
static tw_stime simplewan_packet_event(
125 126 127 128 129 130
        char* category,
        tw_lpid final_dest_lp,
        uint64_t packet_size,
        int is_pull,
        uint64_t pull_size, /* only used when is_pull == 1 */
        tw_stime offset,
131
        const mn_sched_params *sched_params,
132 133 134 135
        int remote_event_size,
        const void* remote_event,
        int self_event_size,
        const void* self_event,
136
        tw_lpid src_lp,
137 138
        tw_lp *sender,
        int is_last_pckt);
139 140 141 142 143 144
static void simplewan_packet_event_rc(tw_lp *sender);

static void simplewan_packet_event_rc(tw_lp *sender);

static void sw_report_stats();

145 146 147 148
static tw_lpid sw_find_local_device(
        const char * annotation,
        int ignore_annotations,
        tw_lp *sender);
149 150 151 152

/* data structure for model-net statistics */
struct model_net_method simplewan_method =
{
153
    .mn_configure = sw_configure,
154 155
    .model_net_method_packet_event = simplewan_packet_event,
    .model_net_method_packet_event_rc = simplewan_packet_event_rc,
156 157
    .model_net_method_recv_msg_event = NULL,
    .model_net_method_recv_msg_event_rc = NULL,
158 159 160
    .mn_get_lp_type = sw_get_lp_type,
    .mn_get_msg_sz = sw_get_msg_sz,
    .mn_report_stats = sw_report_stats,
161
    .model_net_method_find_local_device = sw_find_local_device,
162 163
    .mn_collective_call = simple_wan_collective,
    .mn_collective_call_rc = simple_wan_collective_rc  
164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183
};

static void sw_init(
    sw_state * ns,
    tw_lp * lp);
static void sw_event(
    sw_state * ns,
    tw_bf * b,
    sw_message * m,
    tw_lp * lp);
static void sw_rev_event(
    sw_state * ns,
    tw_bf * b,
    sw_message * m,
    tw_lp * lp);
static void sw_finalize(
    sw_state * ns,
    tw_lp * lp);

tw_lptype sw_lp = {
184 185 186 187 188 189 190
    (init_f) sw_init,
    (pre_run_f) NULL,
    (event_f) sw_event,
    (revent_f) sw_rev_event,
    (final_f) sw_finalize,
    (map_f) codes_mapping,
    sizeof(sw_state),
191 192
};

193
static tw_stime rate_to_ns(uint64_t bytes, double MB_p_s);
194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214
static void handle_msg_ready_rev_event(
    sw_state * ns,
    tw_bf * b,
    sw_message * m,
    tw_lp * lp);
static void handle_msg_ready_event(
    sw_state * ns,
    tw_bf * b,
    sw_message * m,
    tw_lp * lp);
static void handle_msg_start_rev_event(
    sw_state * ns,
    tw_bf * b,
    sw_message * m,
    tw_lp * lp);
static void handle_msg_start_event(
    sw_state * ns,
    tw_bf * b,
    sw_message * m,
    tw_lp * lp);

215 216 217 218 219 220 221 222 223 224 225 226 227
/* collective network calls */
static void simple_wan_collective()
{
/* collectives not supported */
    return;
}

static void simple_wan_collective_rc()
{
/* collectives not supported */
   return;
}

228 229 230 231 232 233 234 235 236 237 238 239 240 241
/* returns pointer to LP information for simplewan module */
static const tw_lptype* sw_get_lp_type()
{
    return(&sw_lp);
}

/* returns number of bytes that the simplewan module will consume in event
 * messages
 */
static int sw_get_msg_sz(void)
{
    return(sizeof(sw_message));
}

242
static double * parse_mat(char * buf, int *nvals_first, int *nvals_total, int is_tri_mat){
243 244 245 246 247 248 249
    int bufn = 128;
    double *vals = malloc(bufn*sizeof(double));

    *nvals_first = 0;
    *nvals_total = 0;

    /* parse the files by line */ 
250
    int line_ct, line_ct_prev = 0;
251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269
    char * line_save;
    char * line = strtok_r(buf, "\r\n", &line_save);
    while (line != NULL){
        line_ct = 0;
        char * tok_save;
        char * tok = strtok_r(line, " \t", &tok_save);
        while (tok != NULL){
            if (line_ct + *nvals_total >= bufn){
                bufn<<=1;
                vals = realloc(vals, bufn*sizeof(double));
            }
            vals[line_ct+*nvals_total] = atof(tok);
            line_ct++;
            tok = strtok_r(NULL, " \t", &tok_save);
        }
        /* first line check - number of tokens = the matrix dim */
        if (*nvals_first == 0) {
            *nvals_first = line_ct;
        }
270
        else if (is_tri_mat && line_ct != line_ct_prev-1){
271 272 273
            fprintf(stderr, "ERROR: tokens in line don't match triangular matrix format\n");
            exit(1);
        }
274 275 276 277
        else if (!is_tri_mat && line_ct != line_ct_prev){
            fprintf(stderr, "ERROR: tokens in line don't match square matrix format\n");
            exit(1);
        }
278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303
        *nvals_total += line_ct;
        line_ct_prev = line_ct;
        line = strtok_r(NULL, "\r\n", &line_save);
    }
    return vals;
}

static void fill_tri_mat(int N, double *mat, double *tri){
    int i, j, p = 0;
    /* first fill in triangular mat entries */
    for (i = 0; i < N; i++){
        double *row = mat + i*N;
        row[i] = 0.0;
        for (j = i+1; j < N; j++){
            row[j] = tri[p++];
        }
    }
    /* now fill in remaining entries (basically a transpose) */
    for (i = 1; i < N; i++){
        for (j = 0; j < i; j++){
            mat[i*N+j] = mat[j*N+i];
        }
    }
}

/* lets caller specify model parameters to use */
304 305 306 307
static void sw_set_params(
        const char      * startup_fname,
        const char      * bw_fname,
        simplewan_param * params){
308
    long int fsize_s, fsize_b;
309 310
    /* TODO: make this a run-time option */
    int is_tri_mat = 0;
311 312 313 314

    /* slurp the files */
    FILE *sf = fopen(startup_fname, "r");
    FILE *bf = fopen(bw_fname, "r");
315 316 317 318
    if (!sf)
        tw_error(TW_LOC, "simplewan: unable to open %s", startup_fname);
    if (!bf)
        tw_error(TW_LOC, "simplewan: unable to open %s", bw_fname);
319 320 321 322 323 324 325 326 327 328
    fseek(sf, 0, SEEK_END);
    fsize_s = ftell(sf);
    fseek(sf, 0, SEEK_SET);
    fseek(bf, 0, SEEK_END);
    fsize_b = ftell(bf);
    fseek(bf, 0, SEEK_SET);
    char *sbuf = malloc(fsize_s+1);
    sbuf[fsize_s] = '\0';
    char *bbuf = malloc(fsize_b+1);
    bbuf[fsize_b] = '\0';
329 330
    assert(fread(sbuf, 1, fsize_s, sf) == fsize_s);
    assert(fread(bbuf, 1, fsize_b, bf) == fsize_b);
331 332 333 334 335
    fclose(sf);
    fclose(bf);

    int nvals_first_s, nvals_first_b, nvals_total_s, nvals_total_b;

336 337 338
    double *startup_tmp = parse_mat(sbuf, &nvals_first_s, 
            &nvals_total_s, is_tri_mat);
    double *bw_tmp = parse_mat(bbuf, &nvals_first_b, &nvals_total_b, is_tri_mat);
339 340 341

    /* convert tri mat into a regular mat */
    assert(nvals_first_s == nvals_first_b);
342
    params->mat_len = nvals_first_s + ((is_tri_mat) ? 1 : 0);
343
    if (is_tri_mat){
344 345 346 347 348 349
        params->net_startup_ns_table = 
            malloc(params->mat_len*params->mat_len*sizeof(double));
        params->net_bw_mbps_table = 
            malloc(params->mat_len*params->mat_len*sizeof(double));
        fill_tri_mat(params->mat_len, params->net_startup_ns_table, startup_tmp);
        fill_tri_mat(params->mat_len, params->net_bw_mbps_table, bw_tmp);
350 351 352 353
        free(startup_tmp);
        free(bw_tmp);
    }
    else{
354 355
        params->net_startup_ns_table = startup_tmp;
        params->net_bw_mbps_table = bw_tmp;
356
    }
357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373

    /* done */
}

/* report network statistics */
static void sw_report_stats()
{
   /* TODO: Do we have some simplewan statistics to report like we have for torus and dragonfly? */
   return;
}
static void sw_init(
    sw_state * ns,
    tw_lp * lp)
{
    uint32_t h1 = 0, h2 = 0;
    memset(ns, 0, sizeof(*ns));

374
    bj_hashlittle2(LP_METHOD_NM, strlen(LP_METHOD_NM), &h1, &h2);
375 376 377
    sw_magic = h1+h2;
    /* printf("\n sw_magic %d ", sw_magic); */

378 379 380 381 382 383
    ns->anno = codes_mapping_get_annotation_by_lpid(lp->gid);
    if (ns->anno == NULL)
        ns->params = &all_params[num_params-1];
    else{
        int id = configuration_get_annotation_index(ns->anno, anno_map);
        ns->params = &all_params[id];
384 385
    }

386 387 388
    /* inititalize global logical ID w.r.t. annotation */
    ns->id = codes_mapping_get_lp_relative_id(lp->gid, 0, 1);

389
    /* all devices are idle to begin with */
390 391 392 393
    ns->send_next_idle = malloc(ns->params->num_lps * 
            sizeof(ns->send_next_idle));
    ns->recv_next_idle = malloc(ns->params->num_lps * 
            sizeof(ns->recv_next_idle));
394 395
    tw_stime st = tw_now(lp);
    int i;
396
    for (i = 0; i < ns->params->num_lps; i++){
397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421
        ns->send_next_idle[i] = st;
        ns->recv_next_idle[i] = st;
    }

    for (i = 0; i < CATEGORY_MAX; i++){
        ns->idle_times_cat[i].send_next_idle_all = 0.0;
        ns->idle_times_cat[i].send_prev_idle_all = 0.0;
        ns->idle_times_cat[i].recv_next_idle_all = 0.0;
        ns->idle_times_cat[i].recv_prev_idle_all = 0.0;
        ns->idle_times_cat[i].category[0] = '\0';
    }

    return;
}

static void sw_event(
    sw_state * ns,
    tw_bf * b,
    sw_message * m,
    tw_lp * lp)
{
    assert(m->magic == sw_magic);

    switch (m->event_type)
    {
422
        case SW_MSG_START:
423 424
            handle_msg_start_event(ns, b, m, lp);
            break;
425
        case SW_MSG_READY:
426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443
            handle_msg_ready_event(ns, b, m, lp);
            break;
        default:
            assert(0);
            break;
    }
}

static void sw_rev_event(
    sw_state * ns,
    tw_bf * b,
    sw_message * m,
    tw_lp * lp)
{
    assert(m->magic == sw_magic);

    switch (m->event_type)
    {
444
        case SW_MSG_START:
445 446
            handle_msg_start_rev_event(ns, b, m, lp);
            break;
447
        case SW_MSG_READY:
448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483
            handle_msg_ready_rev_event(ns, b, m, lp);
            break;
        default:
            assert(0);
            break;
    }

    return;
}

static void sw_finalize(
    sw_state * ns,
    tw_lp * lp)
{
    /* first need to add last known active-range times (they aren't added 
     * until afterwards) */ 
    int i;
    for (i = 0; 
            i < CATEGORY_MAX && strlen(ns->idle_times_cat[i].category) > 0; 
            i++){
        category_idles *id = ns->idle_times_cat + i;
        mn_stats       *st = ns->sw_stats_array + i;
        st->send_time += id->send_next_idle_all - id->send_prev_idle_all;
        st->recv_time += id->recv_next_idle_all - id->recv_prev_idle_all;
    }

    model_net_print_stats(lp->gid, &ns->sw_stats_array[0]);
    return;
}

int sw_get_magic()
{
  return sw_magic;
}

/* convert MiB/s and bytes to ns */
484
static tw_stime rate_to_ns(uint64_t bytes, double MB_p_s)
485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517
{
    tw_stime time;

    /* bytes to MB */
    time = ((double)bytes)/(1024.0*1024.0);
    /* MB to s */
    time = time / MB_p_s;
    /* s to ns */
    time = time * 1000.0 * 1000.0 * 1000.0;

    return(time);
}

/* reverse computation for msg ready event */
static void handle_msg_ready_rev_event(
    sw_state * ns,
    tw_bf * b,
    sw_message * m,
    tw_lp * lp)
{
    struct mn_stats* stat;
    category_idles * idles;

    stat = model_net_find_stats(m->category, ns->sw_stats_array);
    stat->recv_count--;
    stat->recv_bytes -= m->net_msg_size_bytes;
    stat->recv_time = m->recv_time_saved;

    ns->recv_next_idle[m->src_mn_rel_id] = m->recv_next_idle_saved;
    idles = sw_get_category_idles(m->category, ns->idle_times_cat);
    idles->recv_next_idle_all = m->recv_next_idle_all_saved;
    idles->recv_prev_idle_all = m->recv_prev_idle_all_saved;

518
    if (m->event_size_bytes && m->is_pull){
519
        int net_id = model_net_get_id(LP_METHOD_NM);
520 521 522
        model_net_event_rc(net_id, lp, m->pull_size);
    }

523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539
    return;
}

/* handler for msg ready event.  This indicates that a message is available
 * to recv, but we haven't checked to see if the recv queue is available yet
 */
static void handle_msg_ready_event(
    sw_state * ns,
    tw_bf * b,
    sw_message * m,
    tw_lp * lp)
{
    tw_stime recv_queue_time = 0;
    tw_event *e_new;
    struct mn_stats* stat;

    /* get source->me network stats */
540 541 542 543
    double bw = sw_get_table_ent(m->src_mn_rel_id, ns->id,
            ns->params->num_lps, ns->params->net_bw_mbps_table);
    double startup = sw_get_table_ent(m->src_mn_rel_id, ns->id,
            ns->params->num_lps, ns->params->net_startup_ns_table);
544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573
    if (bw <= 0.0 || startup < 0.0){
        fprintf(stderr, 
                "Invalid link from Rel. id %d to LP %lu (rel. id %d)\n", 
                m->src_mn_rel_id, lp->gid, ns->id);
        abort();
    }

    /* are we available to recv the msg? */
    /* were we available when the transmission was started? */
    if(ns->recv_next_idle[m->src_mn_rel_id] > tw_now(lp))
        recv_queue_time += 
            ns->recv_next_idle[m->src_mn_rel_id] - tw_now(lp);

    /* calculate transfer time based on msg size and bandwidth */
    recv_queue_time += rate_to_ns(m->net_msg_size_bytes, bw);

    /* bump up input queue idle time accordingly */
    m->recv_next_idle_saved = ns->recv_next_idle[m->src_mn_rel_id];
    ns->recv_next_idle[m->src_mn_rel_id] = recv_queue_time + tw_now(lp);

    /* get stats, save state (TODO: smarter save state than param dump?)  */
    stat = model_net_find_stats(m->category, ns->sw_stats_array);
    category_idles *idles = 
        sw_get_category_idles(m->category, ns->idle_times_cat);
    stat->recv_count++;
    stat->recv_bytes += m->net_msg_size_bytes;
    m->recv_time_saved = stat->recv_time;
    m->recv_next_idle_all_saved = idles->recv_next_idle_all;
    m->recv_prev_idle_all_saved = idles->recv_prev_idle_all;

574
#if SIMPLEWAN_DEBUG
575 576 577 578 579 580
    printf("%d: from_id:%d now: %8.3lf next_idle_recv: %8.3lf\n",
            ns->id, m->src_mn_rel_id,
            tw_now(lp), ns->recv_next_idle[m->src_mn_rel_id]);
    printf("%d: BEFORE all_idles_recv %8.3lf %8.3lf\n",
            ns->id,
            idles->recv_prev_idle_all, idles->recv_next_idle_all);
581 582
#endif
    
583 584 585 586 587 588 589 590 591 592 593 594 595

    /* update global idles, recv time */
    if (tw_now(lp) > idles->recv_next_idle_all){
        /* there was an idle period between last idle and now */
        stat->recv_time += 
            idles->recv_next_idle_all - idles->recv_prev_idle_all;
        idles->recv_prev_idle_all = tw_now(lp); 
    }
    if (ns->recv_next_idle[m->src_mn_rel_id] > idles->recv_next_idle_all){
        /* extend the active period (active until at least this request) */
        idles->recv_next_idle_all = ns->recv_next_idle[m->src_mn_rel_id];
    }

596
#if SIMPLEWAN_DEBUG
597 598 599 600 601 602 603 604
    printf("%d: AFTER  all_idles_recv %8.3lf %8.3lf",
            ns->id, idles->recv_prev_idle_all, idles->recv_next_idle_all);
    if (m->event_size_bytes>0){
        printf(" - with event\n");
    }
    else{
        printf(" - without event\n");
    }
605
#endif
606 607 608 609

    /* copy only the part of the message used by higher level */
    if(m->event_size_bytes)
    {
610 611 612
        //char* tmp_ptr = (char*)m;
        //tmp_ptr += sw_get_msg_sz();
        void *tmp_ptr = model_net_method_get_edata(SIMPLEWAN, m);
613
        if (m->is_pull){
614
            int net_id = model_net_get_id(LP_METHOD_NM);
615 616 617 618 619 620
            model_net_event(net_id, m->category, m->src_gid, m->pull_size,
                    recv_queue_time, m->event_size_bytes, tmp_ptr, 0, NULL, lp);
        }
        else{
            /* schedule event to final destination for when the recv is complete */
            e_new = tw_event_new(m->final_dest_gid, recv_queue_time, lp);
621
            void *m_new = tw_event_data(e_new);
622 623 624
            memcpy(m_new, tmp_ptr, m->event_size_bytes);
            tw_event_send(e_new);
        }
625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669
    }

    return;
}

/* reverse computation for msg start event */
static void handle_msg_start_rev_event(
    sw_state * ns,
    tw_bf * b,
    sw_message * m,
    tw_lp * lp)
{
    if(m->local_event_size_bytes > 0)
    {
        codes_local_latency_reverse(lp);
    }

    mn_stats* stat;
    stat = model_net_find_stats(m->category, ns->sw_stats_array);
    stat->send_count--;
    stat->send_bytes -= m->net_msg_size_bytes;
    stat->send_time = m->send_time_saved;

    category_idles *idles = 
        sw_get_category_idles(m->category, ns->idle_times_cat);
    ns->send_next_idle[m->dest_mn_rel_id] = m->send_next_idle_saved;
    idles->send_next_idle_all = m->send_next_idle_all_saved;
    idles->send_prev_idle_all = m->send_prev_idle_all_saved;

    return;
}

/* handler for msg start event; this indicates that the caller is trying to
 * transmit a message through this NIC
 */
static void handle_msg_start_event(
    sw_state * ns,
    tw_bf * b,
    sw_message * m,
    tw_lp * lp)
{
    tw_event *e_new;
    sw_message *m_new;
    tw_stime send_queue_time = 0;
    mn_stats* stat;
670
    int mapping_rep_id, mapping_offset, dummy;
671
    tw_lpid dest_id;
672
    char lp_group_name[MAX_NAME_LENGTH];
673 674 675 676
    int total_event_size;
    int dest_rel_id;
    double bw, startup;

677 678
    total_event_size = model_net_get_msg_sz(SIMPLEWAN) + m->event_size_bytes +
        m->local_event_size_bytes;
679

680 681
    codes_mapping_get_lp_info(m->final_dest_gid, lp_group_name, &dummy,
            NULL, &dummy, NULL, &mapping_rep_id, &mapping_offset);
682
    codes_mapping_get_lp_id(lp_group_name, LP_CONFIG_NM, ns->anno, 0,
683 684
            mapping_rep_id, mapping_offset, &dest_id); 
    dest_rel_id = codes_mapping_get_lp_relative_id(dest_id, 0, 0);
685 686 687
    m->dest_mn_rel_id = dest_rel_id;

    /* grab the link params */
688 689 690 691
    bw = sw_get_table_ent(ns->id, dest_rel_id,
            ns->params->num_lps, ns->params->net_bw_mbps_table);
    startup = sw_get_table_ent(ns->id, dest_rel_id,
            ns->params->num_lps, ns->params->net_startup_ns_table);
692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723
    if (bw <= 0.0 || startup < 0.0){
        fprintf(stderr, 
                "Invalid link from LP %lu (rel. id %d) to LP %lu (rel. id %d)\n", 
                lp->gid, ns->id, dest_id, dest_rel_id);
        abort();
    }

    /* calculate send time stamp */
    send_queue_time = startup; /* net msg startup cost */
    /* bump up time if the NIC send queue isn't idle right now */
    if(ns->send_next_idle[dest_rel_id] > tw_now(lp))
        send_queue_time += ns->send_next_idle[dest_rel_id] - tw_now(lp);

    /* move the next idle time ahead to after this transmission is
     * _complete_ from the sender's perspective 
     */ 
    m->send_next_idle_saved = ns->send_next_idle[dest_rel_id];
    ns->send_next_idle[dest_rel_id] = send_queue_time + tw_now(lp) +
        rate_to_ns(m->net_msg_size_bytes, bw);

    /* get stats, save state (TODO: smarter save state than param dump?)  */
    stat = model_net_find_stats(m->category, ns->sw_stats_array);
    category_idles *idles = 
        sw_get_category_idles(m->category, ns->idle_times_cat);
    stat->send_count++;
    stat->send_bytes += m->net_msg_size_bytes;
    m->send_time_saved = stat->send_time;
    m->send_next_idle_all_saved = idles->send_next_idle_all;
    m->send_prev_idle_all_saved = idles->send_prev_idle_all;
    if(stat->max_event_size < total_event_size)
        stat->max_event_size = total_event_size;

724
#if SIMPLEWAN_DEBUG
725 726 727 728 729
    printf("%d: to_id:%d now: %8.3lf next_idle_send: %8.3lf\n",
            ns->id, dest_rel_id,
            tw_now(lp), ns->send_next_idle[dest_rel_id]);
    printf("%d: BEFORE all_idles_send %8.3lf %8.3lf\n",
            ns->id, idles->send_prev_idle_all, idles->send_next_idle_all);
730
#endif
731 732 733 734 735 736 737 738 739 740 741 742

    /* update global idles, send time */
    if (tw_now(lp) > idles->send_next_idle_all){
        /* there was an idle period between last idle and now */
        stat->send_time += idles->send_next_idle_all - idles->send_prev_idle_all;
        idles->send_prev_idle_all = tw_now(lp); 
    }
    if (ns->send_next_idle[dest_rel_id] > idles->send_next_idle_all){
        /* extend the active period (active until at least this request) */
        idles->send_next_idle_all = ns->send_next_idle[dest_rel_id];
    }

743
#if SIMPLEWAN_DEBUG
744 745 746 747 748 749 750 751
    printf("%d: AFTER  all_idles_send %8.3lf %8.3lf",
            ns->id, idles->send_prev_idle_all, idles->send_next_idle_all);
    if (m->local_event_size_bytes>0){
        printf(" - with local event\n");
    }
    else{
        printf(" - without local event\n");
    }
752
#endif
753 754 755

    /* create new event to send msg to receiving NIC */
//    printf("\n msg start sending to %d ", dest_id);
756 757 758 759 760
    void *m_data;
    //e_new = tw_event_new(dest_id, send_queue_time, lp);
    //m_new = tw_event_data(e_new);
    e_new = model_net_method_event_new(dest_id, send_queue_time, lp,
            SIMPLEWAN, (void**)&m_new, &m_data);
761 762 763 764

    /* copy entire previous message over, including payload from user of
     * this module
     */
765 766 767 768 769 770 771
    //memcpy(m_new, m, m->event_size_bytes + sw_get_msg_sz());
    memcpy(m_new, m, sizeof(sw_message));
    if (m->event_size_bytes){
        memcpy(m_data, model_net_method_get_edata(SIMPLEWAN, m),
                m->event_size_bytes);
    }
    m_new->event_type = SW_MSG_READY;
772 773 774 775 776 777 778 779 780
    m_new->src_mn_rel_id = ns->id;
    
    tw_event_send(e_new);

    /* if there is a local event to handle, then create an event for it as
     * well
     */
    if(m->local_event_size_bytes > 0)
    {
781
        //char* local_event;
782 783 784 785

        e_new = tw_event_new(m->src_gid, send_queue_time+codes_local_latency(lp), lp);
        m_new = tw_event_data(e_new);

786 787 788 789
        void * m_loc = (char*) model_net_method_get_edata(SIMPLEWAN, m) +
            m->event_size_bytes;
         //local_event = (char*)m;
         //local_event += sw_get_msg_sz() + m->event_size_bytes;         	 
790
        /* copy just the local event data over */
791
        memcpy(m_new, m_loc, m->local_event_size_bytes);
792 793 794 795 796 797 798 799 800 801
        tw_event_send(e_new);
    }
    return;
}

/* Model-net function calls */

/*This method will serve as an intermediate layer between simplewan and modelnet. 
 * It takes the packets from modelnet layer and calls underlying simplewan methods*/
static tw_stime simplewan_packet_event(
802 803 804 805 806 807
        char* category,
        tw_lpid final_dest_lp,
        uint64_t packet_size,
        int is_pull,
        uint64_t pull_size, /* only used when is_pull == 1 */
        tw_stime offset,
808
        const mn_sched_params *sched_params,
809 810 811 812
        int remote_event_size,
        const void* remote_event,
        int self_event_size,
        const void* self_event,
813
        tw_lpid src_lp,
814 815
        tw_lp *sender,
        int is_last_pckt)
816 817 818 819 820 821 822 823
{
     tw_event * e_new;
     tw_stime xfer_to_nic_time;
     sw_message * msg;
     char* tmp_ptr;

     xfer_to_nic_time = codes_local_latency(sender);

824
#if SIMPLEWAN_DEBUG
825
    printf("%lu: final %lu packet sz %d remote sz %d self sz %d is_last_pckt %d latency %lf\n",
826
            (src_lp - 1) / 2, final_dest_lp, packet_size, 
827 828
            remote_event_size, self_event_size, is_last_pckt,
            xfer_to_nic_time+offset);
829
#endif
830

831 832
     e_new = model_net_method_event_new(sender->gid, xfer_to_nic_time+offset,
             sender, SIMPLEWAN, (void**)&msg, (void**)&tmp_ptr);
833 834
     strcpy(msg->category, category);
     msg->final_dest_gid = final_dest_lp;
835
     msg->src_gid = src_lp;
836 837 838 839
     msg->magic = sw_get_magic();
     msg->net_msg_size_bytes = packet_size;
     msg->event_size_bytes = 0;
     msg->local_event_size_bytes = 0;
840
     msg->event_type = SW_MSG_START;
841 842
     msg->is_pull = is_pull;
     msg->pull_size = pull_size;
843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865

    //printf("\n Sending to LP %d msg magic %d ", (int)dest_id, sw_get_magic()); 
     /*Fill in simplewan information*/     
     if(is_last_pckt) /* Its the last packet so pass in remote event information*/
      {
       if(remote_event_size)
	 {
           msg->event_size_bytes = remote_event_size;
           memcpy(tmp_ptr, remote_event, remote_event_size);
           tmp_ptr += remote_event_size;
	 }
       if(self_event_size)
       {
	   msg->local_event_size_bytes = self_event_size;
	   memcpy(tmp_ptr, self_event, self_event_size);
	   tmp_ptr += self_event_size;
       }
      // printf("\n Last packet size: %d ", sw_get_msg_sz() + remote_event_size + self_event_size);
      }
     tw_event_send(e_new);
     return xfer_to_nic_time;
}

866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912
static void sw_read_config(const char * anno, simplewan_param *p){
    char startup_file[MAX_NAME_LENGTH];
    char bw_file[MAX_NAME_LENGTH];
    int rc;
    rc = configuration_get_value_relpath(&config, "PARAMS", 
            "net_startup_ns_file", anno, startup_file, MAX_NAME_LENGTH);
    if (rc == 0){
        if (anno == NULL)
            tw_error(TW_LOC,
                    "simplewan: unable to read PARAMS:net_startup_ns_file");
        else
            tw_error(TW_LOC,
                    "simplewan: unable to read PARAMS:net_startup_ns_file@%s",
                    anno);
    }
    rc = configuration_get_value_relpath(&config, "PARAMS", "net_bw_mbps_file",
            anno, bw_file, MAX_NAME_LENGTH);
    if (rc == 0){
        if (anno == NULL)
            tw_error(TW_LOC,
                    "simplewan: unable to read PARAMS:net_bw_mbps_file");
        else
            tw_error(TW_LOC,
                    "simplewan: unable to read PARAMS:net_bw_mbps_file@%s",
                    anno);
    }
    p->num_lps = codes_mapping_get_lp_count(NULL, 0,
            LP_CONFIG_NM, anno, 0);
    sw_set_params(startup_file, bw_file, p);
    if (p->mat_len != p->num_lps){
        tw_error(TW_LOC, "Simplewan config matrix doesn't match the "
                "number of simplewan LPs (%d vs. %d)\n",
                p->mat_len, p->num_lps);
    }
}

static void sw_configure(){
    anno_map = codes_mapping_get_lp_anno_map(LP_CONFIG_NM);
    assert(anno_map);
    num_params = anno_map->num_annos + (anno_map->has_unanno_lp > 0);
    all_params = malloc(num_params * sizeof(*all_params));
    for (uint64_t i = 0; i < anno_map->num_annos; i++){
        sw_read_config(anno_map->annotations[i], &all_params[i]);
    }
    if (anno_map->has_unanno_lp > 0){
        sw_read_config(NULL, &all_params[anno_map->num_annos]);
    }
913 914 915 916 917 918 919 920
}

static void simplewan_packet_event_rc(tw_lp *sender)
{
    codes_local_latency_reverse(sender);
    return;
}

921 922 923 924
static tw_lpid sw_find_local_device(
        const char * annotation,
        int ignore_annotations,
        tw_lp *sender)
925
{
926 927
     char lp_group_name[MAX_NAME_LENGTH];
     int mapping_rep_id, mapping_offset, dummy;
928 929
     tw_lpid dest_id;

930 931
     // TODO: don't ignore annotation
     codes_mapping_get_lp_info(sender->gid, lp_group_name, &dummy, NULL, &dummy, NULL, &mapping_rep_id, &mapping_offset);
932 933
     codes_mapping_get_lp_id(lp_group_name, LP_CONFIG_NM, annotation,
             ignore_annotations, mapping_rep_id, mapping_offset, &dest_id);
934 935 936 937

    return(dest_id);
}

938 939 940 941 942 943 944
static double sw_get_table_ent(
        int      from_id, 
        int      to_id,
        int      num_lps,
        double * table){
    // TODO: if a tri-matrix, then change the addressing
    return table[from_id * num_lps + to_id]; 
945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981
}

/* category lookup (more or less copied from model_net_find_stats) */
static category_idles* sw_get_category_idles(
        char * category, category_idles *idles){
    int i;
    int new_flag = 0;
    int found_flag = 0;

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

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

/*
 * Local variables:
 *  c-indent-level: 4
 *  c-basic-offset: 4
 * End:
 *
 * vim: ft=c ts=8 sts=4 sw=4 expandtab
 */