Merging Xu's multi-application workload with MPI Sim layer, enabling...

Merging Xu's multi-application workload with MPI Sim layer, enabling time-stepped series data output for the MPI traces (single and multiple applications), updating documentation

src/network-workloads/README_traces.txt

@@ -22,7 +22,7 @@
 
    http://portal.nersc.gov/project/CAL/designforward.htm
 
------------------ RUNNING CODES MPI SIMULATION LAYER -----------------------
+--- RUNNING CODES MPI SIMULATION LAYER (DEFAULT JOB ALLOCATION, SINGLE WORKLOAD)-------------
 6- Download and untar the DUMPI AMG application trace for 216 MPI ranks using the following download link:
 
 wget
@@ -42,7 +42,12 @@ src/network-workloads/conf/modelnet-mpi-test-dfly-amg-216.conf)
 
   The simulation runs in ROSS serial, conservative and optimistic modes.
 
-   Note: Dragonfly and torus networks may have more number of nodes in the network than the number network traces (Some network nodes will only pass messages and they will not end up loading the traces). Thats why --num_net_traces argument is used to specify exact number of traces available in the DUMPI directory if there is a mis-match between number of network nodes and traces.
+   Note: Dragonfly and torus networks may have more number of nodes in the
+   network than the number network traces (Some network nodes will only pass
+   messages and they will not end up loading the traces). Thats why
+   --num_net_traces argument is used to specify exact number of traces
+   available in the DUMPI directory if there is a mis-match between number of
+   network nodes and traces.
 
 10- Running the simulation in optimistic mode 
     
@@ -52,3 +57,48 @@ src/network-workloads/conf/modelnet-mpi-test-dfly-amg-216.conf)
     --workload_file=/projects/radix-io/mubarak/df_traces/directory/dumpi-2014.03.03.15.09.03- 
     -- src/network-workloads//conf/modelnet-mpi-test-dfly-amg-216.conf 
 
+--- RUNNING MPI SIMULATION LAYER WITH MULTIPLE WORKLOADS --------
+11- Generate job allocation file (random or contiguous) using python scripts.
+
+Allocation options 
+
+- Random allocation assigns a set of randomly selected network nodes to each
+job. 
+- Contiguous allocation assigns a set of contiguous network nodes to the
+jobs. 
+
+See codes/allocation_gen/README for instructions on how to generate job
+allocation file using python. Example allocation files are in
+src/network-workloads/conf/allocation-rand.conf, allocation-cont.conf.
+
+12- Run the simulation with multiple job allocations
+
+./src/network-workloads//model-net-mpi-replay --sync=1
+--workload_conf_file=workloads.conf
+--alloc_file=../src/network-workloads/conf/allocation-rand.conf
+--workload_type="dumpi" --
+../src/network-workloads/conf/modelnet-mpi-test-dfly-amg-216.conf 
+
+To run in optimistic mode:
+
+mpirun -np 4 ./src/network-workloads//model-net-mpi-replay --sync=3
+--workload_conf_file=workloads.conf --alloc_file=allocation.conf
+--workload_type="dumpi" --
+../src/network-workloads/conf/modelnet-mpi-test-dfly-amg-216.conf
+
+----- sampling and debugging options for MPI Simulation Layer ---- 
+
+Runtime options can be used to enable time-stepped series data of simulation
+with multiple workloads:
+
+--enable_sampling = 1 [Turns on sampling after a specific simulated interval.
+Default sampling interval is 5 millisec and default sampling end time is 3
+secs.]
+
+--lp-io-dir-dir-name [Turns on end of simulation statistics for dragonfly network model]
+
+--lp-io-use-suffix [Output to a unique directory ]
+
+--enable_mpi_debug=1 prints the details of the MPI operations being simulated. Enabling
+debug mode can display a lot of print statements!.
+

src/network-workloads/allocation_gen/README

+[Notes from Xu Yang]
+
+To generate the allocation file:
+
+python listgen.py config_alloc.conf
+
+There are three lines in the config_alloc.conf file. 
+1st line: allocation policy, you can use CONT, rand, stripe etc.
+2nd line: the total terminal number of nodes in the system, if you use 8-router
+dragonfly, this number would be 1056
+3rd line: the number of ranks of each application. 
+
+

src/network-workloads/allocation_gen/allocation.conf

+0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 
+216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 
+341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 

src/network-workloads/allocation_gen/config_alloc.conf

+CONT
+1056
+216 125 100

src/network-workloads/allocation_gen/listgen.py

+import sys
+import random
+alloc_file = 'allocation.conf'
+
+def contiguous_alloc(job_ranks, total_nodes):
+    f = open(alloc_file,'w')
+    start=0
+    for num_rank in range(len(job_ranks)):
+        for rankid in range(start, start+job_ranks[num_rank]):
+            f.write("%s " % rankid)
+        f.write("\n")
+        start += job_ranks[num_rank]
+    f.closed
+
+def cube_alloc(job_ranks, total_nodes):
+    job_dim = [6,6,6]
+    sys_dim_x = 16
+    sys_dim_y =16
+    sys_dim_z = 8
+    cube = []
+    start = 0
+    for k in range(job_dim[2]):
+        layer = []
+        layer_offset = k*sys_dim_x*sys_dim_y
+        for j in range(job_dim[1]):
+            row_offset = j*sys_dim_z
+            row = []
+            for i in range(job_dim[0]):
+                offset = row_offset+layer_offset
+                row.append(i+offset)
+            layer += row
+        cube += layer
+    print "list length is", len(cube), cube
+
+    f = open('cube_allc_linear.conf','w')
+    for rankid in range(len(cube)):
+        f.write("%s " % cube[rankid])
+    f.write("\n")
+    f.closed
+
+    f = open('cube_allc_random.conf','w')
+    random.shuffle(cube)
+    for rankid in range(len(cube)):
+        f.write("%s " % cube[rankid])
+    f.write("\n")
+    f.closed
+
+
+
+def permeate_alloc(job_ranks, total_nodes):
+    f = open(alloc_file,'w')
+    start=0
+    node_list = range(0, int(total_nodes))
+    random.seed(0)
+    for num_rank in range(len(job_ranks)):
+        permeate_area = job_ranks[num_rank]*4
+        permeate_list = node_list[num_rank*permeate_area: (num_rank+1)*permeate_area]
+        alloc_list = random.sample(permeate_list, job_ranks[num_rank])
+        alloc_list.sort()
+        print "length of alloc list", len(alloc_list), "\n", alloc_list,"\n"
+        for idx in range(len(alloc_list)):
+            f.write("%s " % alloc_list[idx])
+        f.write("\n")
+    f.closed
+
+def random_alloc(job_rank, total_nodes):
+    f = open(alloc_file, 'w')
+    node_list = range(0, int(total_nodes))
+    random.seed(0)
+    for rankid in range(len(job_rank)):
+        alloc_list = random.sample(node_list, job_rank[rankid])
+        node_list = [i for i in node_list if (i not in alloc_list)]
+        print "length of alloc list", len(alloc_list), "\n", alloc_list,"\n"
+        for idx in range(len(alloc_list)):
+            f.write("%s " % alloc_list[idx])
+        f.write("\n")
+    f.closed
+
+
+def stripe_alloc(job_ranks, total_nodes):
+    #print "the num of nodes of each Job", job_ranks
+    f = open(alloc_file,'w')
+    node_list = range(0, int(total_nodes))
+    stripe_size = 2
+    alloc_list = []
+    for num_rank in range(len(job_ranks)):
+    #    print "job id", num_rank
+        num_stripe = 1
+        start = num_rank*stripe_size
+        if(job_ranks[num_rank] % stripe_size != 0):
+            num_stripe = job_ranks[num_rank]/stripe_size+1
+        else:
+            num_stripe = job_ranks[num_rank]/stripe_size
+        tmp_list = []
+        while(num_stripe>0):
+            tmp_list += node_list[start:start+stripe_size]
+            start += len(job_ranks)*stripe_size
+            num_stripe -= 1
+        alloc_list.append(tmp_list)
+
+
+    for job_id in range (len(alloc_list)):
+        tmp = alloc_list[job_id]
+        #print "alloc list for JOB", job_id
+        for rankid in range (job_ranks[job_id]):
+           # print tmp[rankid]
+            f.write("%s " % tmp[rankid])
+        f.write("\n")
+    f.closed
+
+def policy_select(plcy, job_ranks, total_nodes):
+    if plcy == "CONT":
+        print "contiguous alloction!"
+        contiguous_alloc(job_ranks,  total_nodes)
+    elif plcy == "rand":
+        print "random allocation!"
+        random_alloc(job_ranks, total_nodes)
+    elif plcy == "STRIPE":
+        print "stripe allcation!"
+        stripe_alloc(job_ranks, total_nodes)
+    elif plcy == "PERMEATE":
+        print "permeate allocation!"
+        permeate_alloc(job_ranks, total_nodes)
+    elif plcy == "CUBE":
+        print "cube allocation!"
+        cube_alloc(job_ranks, total_nodes)
+    else:
+        print "NOT Supported yet!"
+
+
+if __name__ == "__main__":
+    f = open(sys.argv[1], "r")
+    array = []
+
+    for line in f:
+        for number in line.split():
+            array.append(number);
+
+    f.close()
+    alloc_plcy = array.pop(0)
+    total_nodes = array.pop(0)
+    print alloc_plcy
+    array = map(int, array)
+    print array
+    policy_select(alloc_plcy, array, total_nodes)
+

src/network-workloads/conf/allocation-cont.conf

+0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 
+216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 
+341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 

src/network-workloads/conf/allocation-random.conf

+891 800 444 273 539 427 827 320 503 616 958 532 297 798 652 264 960 1037 855 952 327 770 949 722 498 106 458 645 964 1020 913 275 850 579 14 760 421 871 705 1 521 916 257 343 919 201 599 251 1021 848 473 84 337 536 985 115 582 746 578 860 570 1017 636 620 469 629 406 607 306 199 197 647 693 94 925 975 889 948 974 413 744 291 857 897 945 622 1002 612 475 697 1052 968 837 86 513 665 892 256 772 123 232 839 351 861 154 736 47 606 564 718 28 670 640 608 390 1035 38 22 1014 195 130 222 845 989 24 449 107 274 233 683 369 190 531 41 1043 210 378 885 969 178 710 61 714 361 630 467 184 432 600 537 328 377 884 591 13 783 354 48 296 253 1006 371 303 379 999 669 655 755 409 437 687 203 353 252 673 399 924 424 741 441 699 49 470 166 557 514 592 797 933 522 329 493 854 198 1055 668 88 766 1042 716 333 225 757 2 868 103 125 685 922 295 1033 105 901 418 85 290 
+472 826 903 137 541 679 357 908 282 19 42 707 571 982 973 939 43 774 724 674 732 927 653 375 544 206 589 9 148 332 801 729 336 623 1041 158 991 283 391 545 285 474 231 50 170 512 69 394 318 404 96 896 462 823 962 331 464 684 411 286 720 870 895 602 359 944 614 63 78 726 59 8 373 492 425 988 552 646 397 345 940 240 738 402 334 60 813 659 844 419 632 111 366 183 39 881 191 134 174 953 500 1050 911 906 135 367 621 805 450 833 980 451 910 603 929 548 188 301 867 810 947 32 132 223 703 
+879 605 756 836 72 90 904 40 227 44 16 869 330 160 150 676 1004 515 926 509 577 692 815 773 1012 761 918 204 534 595 824 484 791 386 580 842 792 654 0 173 501 243 68 841 932 287 398 1005 65 625 122 268 806 56 34 401 479 829 694 648 144 955 389 1026 363 46 447 143 31 965 588 99 516 315 352 734 459 820 565 436 583 307 116 633 569 942 117 834 742 835 796 619 747 477 719 165 713 400 688 410 

src/network-workloads/conf/modelnet-mpi-test-dfly-amg-216.conf

@@ -23,6 +23,6 @@ PARAMS
    local_bandwidth="5.25";
    global_bandwidth="4.7";
    cn_bandwidth="5.25";
-   message_size="560";
+   message_size="576";
    routing="adaptive";
 }

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

@@ -23,7 +23,6 @@
 #define lprintf(_fmt, ...) \
         do {if (CS_LP_DBG) printf(_fmt, __VA_ARGS__);} while (0)
 #define MAX_STATS 65536
-//#define WORKLOAD_LOG 1
 
 char workload_type[128];
 char workload_file[8192];
@@ -78,6 +77,7 @@ static int disable_delay = 0;
 static int enable_sampling = 0;
 static double sampling_interval = 5000000;
 static double sampling_end_time = 3000000000;
+static int enable_debug = 1;
 
 /* MPI_OP_GET_NEXT is for getting next MPI operation when the previous operation completes.
 * MPI_SEND_ARRIVED is issued when a MPI message arrives at its destination (the message is transported by model-net and an event is invoked when it arrives. 
@@ -94,6 +94,7 @@ struct mpi_workload_sample
 {
     /* Sampling data */
     int nw_id;
+    int app_id;
     unsigned long num_sends_sample;
     unsigned long num_bytes_sample;
     unsigned long num_waits_sample;
@@ -211,6 +212,7 @@ struct nw_message
        double msg_send_time;
        int16_t req_id;   
        int tag;
+       int app_id;
        int found_match;
        short wait_completed;
    } fwd;
@@ -371,9 +373,8 @@ static int notify_posted_wait(nw_state* s,
                 assert(wait_elem->num_completed <= wait_elem->count);
                 if(wait_elem->num_completed == wait_elem->count)
                 {
-                 #if WORKLOAD_LOG == 1
-                    fprintf(workload_log, "\n(%lf) MPI WAITALL COMPLETED AT %ld ", tw_now(lp), s->nw_id);
-                #endif
+                    if(enable_debug)
+                        fprintf(workload_log, "\n(%lf) APP ID %d MPI WAITALL COMPLETED AT %ld ", s->app_id, tw_now(lp), s->nw_id);
                     wait_completed = 1;
                 }
            
@@ -471,9 +472,9 @@ static void codes_exec_mpi_wait_all(
         tw_lp* lp, 
         struct codes_workload_op * mpi_op)
 {
-#if WORKLOAD_LOG == 1
-  fprintf(workload_log, "\n MPI WAITALL POSTED AT %ld ", s->nw_id);
-#endif
+  if(enable_debug)
+    fprintf(workload_log, "\n MPI WAITALL POSTED AT %ld ", s->nw_id);
+  
   if(enable_sampling)
   {
     bf->c1 = 1;
@@ -482,13 +483,14 @@ static void codes_exec_mpi_wait_all(
         bf->c2 = 1;
         int indx = s->sampling_indx;
         s->mpi_wkld_samples[indx].nw_id = s->nw_id;
+        s->mpi_wkld_samples[indx].app_id = s->app_id;
         s->mpi_wkld_samples[indx].sample_end_time = s->cur_interval_end;
         s->cur_interval_end += sampling_interval;
         s->sampling_indx++;
     }
     if(s->sampling_indx >= MAX_STATS)
     {
-        struct mpi_workload_sample * tmp = malloc((MAX_STATS + s->max_arr_size) * sizeof(struct mpi_workload_sample));
+        struct mpi_workload_sample * tmp = calloc((MAX_STATS + s->max_arr_size), sizeof(struct mpi_workload_sample));
         memcpy(tmp, s->mpi_wkld_samples, s->sampling_indx);
         free(s->mpi_wkld_samples);
         s->mpi_wkld_samples = tmp;
@@ -836,13 +838,14 @@ static void codes_exec_mpi_send(nw_state* s,
             bf->c1 = 1;
             int indx = s->sampling_indx;
             s->mpi_wkld_samples[indx].nw_id = s->nw_id;
+            s->mpi_wkld_samples[indx].app_id = s->app_id;
             s->mpi_wkld_samples[indx].sample_end_time = s->cur_interval_end;
             s->sampling_indx++;
             s->cur_interval_end += sampling_interval; 
         }
         if(s->sampling_indx >= MAX_STATS)
         {
-            struct mpi_workload_sample * tmp = malloc((MAX_STATS + s->max_arr_size) * sizeof(struct mpi_workload_sample));
+            struct mpi_workload_sample * tmp = calloc((MAX_STATS + s->max_arr_size), sizeof(struct mpi_workload_sample));
             memcpy(tmp, s->mpi_wkld_samples, s->sampling_indx);
             free(s->mpi_wkld_samples);
             s->mpi_wkld_samples = tmp;
@@ -863,6 +866,7 @@ static void codes_exec_mpi_send(nw_state* s,
     local_m.fwd.tag = mpi_op->u.send.tag;
     local_m.fwd.num_bytes = mpi_op->u.send.num_bytes;
     local_m.fwd.req_id = mpi_op->u.send.req_id;
+    local_m.fwd.app_id = s->app_id;
 
     remote_m = local_m;
 	remote_m.msg_type = MPI_SEND_ARRIVED;
@@ -870,15 +874,17 @@ static void codes_exec_mpi_send(nw_state* s,
 	model_net_event(net_id, "test", dest_rank, mpi_op->u.send.num_bytes, 0.0, 
 	    sizeof(nw_message), (const void*)&remote_m, sizeof(nw_message), (const void*)&local_m, lp);
 
-#if WORKLOAD_LOG == 1
-    if(mpi_op->op_type == CODES_WK_ISEND)
-        fprintf(workload_log, "\n (%lf) MPI ISEND SOURCE %ld DEST %ld BYTES %ld ", 
-                tw_now(lp), s->nw_id, mpi_op->u.send.dest_rank, mpi_op->u.send.num_bytes);
-    else
-        fprintf(workload_log, "\n (%lf) MPI SEND SOURCE %ld DEST %ld BYTES %ld ", 
-                tw_now(lp), s->nw_id, mpi_op->u.send.dest_rank, mpi_op->u.send.num_bytes);
-#endif
-
+    if(enable_debug)
+    {
+        if(mpi_op->op_type == CODES_WK_ISEND)
+        {
+            fprintf(workload_log, "\n (%lf) APP %d MPI ISEND SOURCE %ld DEST %ld BYTES %ld ", 
+                    tw_now(lp), s->app_id, s->nw_id, mpi_op->u.send.dest_rank, mpi_op->u.send.num_bytes);
+        }
+        else
+            fprintf(workload_log, "\n (%lf) APP ID %d MPI SEND SOURCE %ld DEST %ld BYTES %ld ", 
+                    tw_now(lp), s->app_id, s->nw_id, mpi_op->u.send.dest_rank, mpi_op->u.send.num_bytes);
+        }
 	/* isend executed, now get next MPI operation from the queue */ 
 	if(mpi_op->op_type == CODES_WK_ISEND)
 	   codes_issue_next_event(lp);
@@ -1008,12 +1014,23 @@ static void update_arrival_queue_rc(nw_state* s,
 /* once an isend operation arrives, the pending receives queue is checked to find out if there is a irecv that has already been posted. If no isend has been posted, */
 static void update_arrival_queue(nw_state* s, tw_bf * bf, nw_message * m, tw_lp * lp)
 {
+    if(s->app_id != m->fwd.app_id)
+        printf("\n Received message for app %d my id %d my rank %d ", 
+                m->fwd.app_id, s->app_id, s->nw_id);
+    assert(s->app_id == m->fwd.app_id);
+
 	m->rc.saved_recv_time = s->recv_time;
     s->num_bytes_recvd += m->fwd.num_bytes;
 
     // send a callback to the sender to increment times
+    // find the global id of the source
+    int global_src_id = m->fwd.src_rank;
+    if(alloc_spec)
+    {
+        global_src_id = get_global_id_of_job_rank(m->fwd.src_rank, s->app_id);
+    }
     tw_event *e_callback =
-        tw_event_new(rank_to_lpid(m->fwd.src_rank),
+        tw_event_new(rank_to_lpid(global_src_id),
                 codes_local_latency(lp), lp);
     nw_message *m_callback = tw_event_data(e_callback);
     m_callback->msg_type = MPI_SEND_ARRIVED_CB;
@@ -1074,16 +1091,14 @@ void nw_test_init(nw_state* s, tw_lp* lp)
  
    memset(s, 0, sizeof(*s));
    s->nw_id = codes_mapping_get_lp_relative_id(lp->gid, 0, 0);
-   s->mpi_wkld_samples = malloc(MAX_STATS * sizeof(struct mpi_workload_sample)); 
+   s->mpi_wkld_samples = calloc(MAX_STATS, sizeof(struct mpi_workload_sample)); 
    s->sampling_indx = 0;
 
    if(!num_net_traces) 
 	num_net_traces = num_net_lps;
-   /* In this case, the LP will not generate any workload related events*/
-   if(s->nw_id >= num_net_traces)
-	    return;
+  
    assert(num_net_traces <= num_net_lps);
-   
+  
    struct codes_jobmap_id lid; 
 
    if(alloc_spec)
@@ -1092,7 +1107,6 @@ void nw_test_init(nw_state* s, tw_lp* lp)
 
         if(lid.job == -1)
         {
-            printf("network LP nw id %d not generating events, lp gid is %ld \n", (int)s->nw_id, lp->gid); 
             s->app_id = -1;
             s->local_rank = -1;
             return;
@@ -1104,19 +1118,26 @@ void nw_test_init(nw_state* s, tw_lp* lp)
        lid.job = 0;
        lid.rank = s->nw_id;
        s->app_id = 0;
+        
+       if(s->nw_id >= num_net_traces)
+	        return;
    }
+   
+   
    if (strcmp(workload_type, "dumpi") == 0){
        strcpy(params_d.file_name, file_name_of_job[lid.job]);
        params_d.num_net_traces = num_traces_of_job[lid.job];
        params = (char*)&params_d;
        s->app_id = lid.job;
        s->local_rank = lid.rank; 
-       //printf("lp global id: %llu, file name: %s, num traces: %d, app id: %d, local id: %d\n", 
-       //        s->nw_id, params_d.file_name, params_d.num_net_traces, s->app_id, s->local_rank);
+//       printf("network LP nw id %d app id %d generating events, lp gid is %ld \n", s->nw_id, s->app_id, lp->gid); 
    }
 
    wrkld_id = codes_workload_load("dumpi-trace-workload", params, s->app_id, s->local_rank);
 
+   INIT_QLIST_HEAD(&s->arrival_queue);
+   INIT_QLIST_HEAD(&s->pending_recvs_queue);
+   INIT_QLIST_HEAD(&s->completed_reqs);
 
    /* Initialize the RC stack */
    rc_stack_create(&s->processed_ops);
@@ -1125,9 +1146,6 @@ void nw_test_init(nw_state* s, tw_lp* lp)
    assert(s->processed_ops != NULL);
    assert(s->matched_reqs != NULL);
 
-   INIT_QLIST_HEAD(&s->arrival_queue);
-   INIT_QLIST_HEAD(&s->pending_recvs_queue);
-   INIT_QLIST_HEAD(&s->completed_reqs);
 
    /* clock starts ticking when the first event is processed */
    s->start_time = tw_now(lp);
@@ -1143,7 +1161,7 @@ void nw_test_init(nw_state* s, tw_lp* lp)
        s->cur_interval_end = sampling_interval;
        if(!g_tw_mynode && !s->nw_id)
        {
-           fprintf(workload_meta_log, "\n mpi_proc_id num_waits "
+           fprintf(workload_meta_log, "\n mpi_proc_id app_id num_waits "
                    " num_sends num_bytes_sent sample_end_time");
        }
    }
@@ -1152,7 +1170,12 @@ void nw_test_init(nw_state* s, tw_lp* lp)
 
 void nw_test_event_handler(nw_state* s, tw_bf * bf, nw_message * m, tw_lp * lp)
 {
-	*(int *)bf = (int)0;
+	if(s->app_id < 0)
+        printf("\n msg type %d ", m->msg_type);
+
+    assert(s->app_id >= 0 && s->local_rank >= 0);
+
+    *(int *)bf = (int)0;
     rc_stack_gc(lp, s->matched_reqs);
     rc_stack_gc(lp, s->processed_ops);
 
@@ -1185,7 +1208,7 @@ void nw_test_event_handler(nw_state* s, tw_bf * bf, nw_message * m, tw_lp * lp)
 
 static void get_next_mpi_operation_rc(nw_state* s, tw_bf * bf, nw_message * m, tw_lp * lp)
 {
-    codes_workload_get_next_rc2(wrkld_id, 0, (int)s->nw_id);
+    codes_workload_get_next_rc2(wrkld_id, s->app_id, s->local_rank);
 
 	if(m->op_type == CODES_WK_END)
     {
@@ -1278,8 +1301,8 @@ static void get_next_mpi_operation(nw_state* s, tw_bf * bf, nw_message * m, tw_l
 {
 		//struct codes_workload_op * mpi_op = malloc(sizeof(struct codes_workload_op));
         struct codes_workload_op mpi_op;
-        codes_workload_get_next(wrkld_id, 0, (int)s->nw_id, &mpi_op);
-      
+        codes_workload_get_next(wrkld_id, s->app_id, s->local_rank, &mpi_op);
+
         m->op_type = mpi_op.op_type;
 
         if(mpi_op.op_type == CODES_WK_END)
@@ -1360,8 +1383,20 @@ void nw_test_finalize(nw_state* s, tw_lp* lp)
     int written = 0;
     if(!s->nw_id)
         written = sprintf(s->output_buf, "# Format <LP ID> <Terminal ID> <Total sends> <Total Recvs> <Bytes sent> <Bytes recvd> <Send time> <Comm. time> <Compute time>");
-	if(s->nw_id < (tw_lpid)num_net_traces)
-	{
+
+    if(alloc_spec == 1)
+    {
+        struct codes_jobmap_id lid;
+        lid = codes_jobmap_to_local_id(s->nw_id, jobmap_ctx);
+     
+        if(lid.job < 0)
+            return;
+    }
+    else
+    {
+        if(s->nw_id >= num_net_traces)
+            return;
+    }
 		int count_irecv = qlist_count(&s->pending_recvs_queue);
         int count_isend = qlist_count(&s->arrival_queue);
 		printf("\n LP %llu unmatched irecvs %d unmatched sends %d Total sends %ld receives %ld collectives %ld delays %ld wait alls %ld waits %ld send time %lf wait %lf", 
@@ -1401,7 +1436,6 @@ void nw_test_finalize(nw_state* s, tw_lp* lp)
 		//printf("\n LP %ld Time spent in communication %llu ", lp->gid, total_time - s->compute_time);
 	    rc_stack_destroy(s->matched_reqs);    
 	    rc_stack_destroy(s->processed_ops);    
-    }
 }
 
 void nw_test_event_handler_rc(nw_state* s, tw_bf * bf, nw_message * m, tw_lp * lp)
@@ -1436,8 +1470,11 @@ const tw_optdef app_opt [] =
 	TWOPT_GROUP("Network workload test"),
     TWOPT_CHAR("workload_type", workload_type, "workload type (either \"scalatrace\" or \"dumpi\")"),
 	TWOPT_CHAR("workload_file", workload_file, "workload file name"),
+	TWOPT_CHAR("alloc_file", alloc_file, "allocation file name"),
+	TWOPT_CHAR("workload_conf_file", workloads_conf_file, "workload config file name"),
 	TWOPT_UINT("num_net_traces", num_net_traces, "number of network traces"),
     TWOPT_UINT("disable_compute", disable_delay, "disable compute simulation"),
+    TWOPT_UINT("enable_mpi_debug", enable_debug, "enable debugging of MPI sim layer (works with sync=1 only)"),
     TWOPT_UINT("sampling_interval", sampling_interval, "sampling interval for MPI operations"),
 	TWOPT_UINT("enable_sampling", enable_sampling, "enable sampling"),
     TWOPT_STIME("sampling_end_time", sampling_end_time, "sampling_end_time"),
@@ -1504,7 +1541,9 @@ int main( int argc, char** argv )
             ref = fscanf(name_file, "%d %s", &num_traces_of_job[i], file_name_of_job[i]); 
             if(ref!=EOF)
             {
-                printf("\n%d traces of app %s \n", num_traces_of_job[i], file_name_of_job[i]);
+                if(enable_debug)
+                    printf("\n%d traces of app %s \n", num_traces_of_job[i], file_name_of_job[i]);
+    
                 num_net_traces += num_traces_of_job[i]; 
                 i++;
             }
@@ -1517,7 +1556,7 @@ int main( int argc, char** argv )
     }
     else
     {
-        assert(num_net_traces > 0);
+        assert(num_net_traces > 0 && strlen(workload_file));
         strcpy(file_name_of_job[0], workload_file);
         num_traces_of_job[0] = num_net_traces;
         alloc_spec = 0;
@@ -1535,16 +1574,17 @@ int main( int argc, char** argv )
    net_id = *net_ids;
    free(net_ids);
 
-#if WORKLOAD_LOG == 1
-   workload_log = fopen("mpi-op-logs", "w+");
-
-   if(!workload_log)