-
Notifications
You must be signed in to change notification settings - Fork 4
/
experiment.py
434 lines (411 loc) · 14.6 KB
/
experiment.py
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
import os
import sys
import params_template as pm
import datetime
import copy
import time
import numpy as np
import os.path
import multiprocessing
# default sl hyper-parameters configuration
sl_config_dict = {"TRAINING_MODE":"SL", "VALUE_NET":False, \
"POLICY_NN_MODEL":None, "VALUE_NN_MODEL":None, "CHECKPOINT_INTERVAL":50, \
"LEARNING_RATE":0.005, "TOT_NUM_STEPS":200, "VAL_INTERVAL":50, \
"NUM_TS_PER_UPDATE":5, "JOB_ORDER_SHUFFLE":True}
NUM_TEST = 5
PARALLELISM = 10
TASK_ID = -1
def replace_params(map, dir):
pm_md = globals().get('pm', None)
train_config = dict()
if pm_md:
train_config = {key: value for key, value in pm_md.__dict__.iteritems() if not (key.startswith('__') or key.startswith('_'))}
f = open(dir+"parameters.py", 'w')
for key, _ in train_config.iteritems():
if key in map.keys():
train_config[key] = map[key]
if isinstance(train_config[key], basestring):
f.write(str(key) + " = " + "'" + str(train_config[key]) + "'" + '\n')
else:
f.write(str(key) + " = " + str(train_config[key])+'\n')
f.close()
def get_config(id, exp_name, test_value):
config = dict()
config["EXPERIMENT_NAME"] = exp_name + "_" + str(test_value)
if id == 1:
config["SCHED_WINDOW_SIZE"] = test_value
config["STATE_DIM"] = (sum([enable for (_,enable) in pm.INPUTS_GATE]), test_value)
config["ACTION_DIM"] = 3 * test_value + pm.SKIP_TS
config["NUM_NEURONS_PER_FCN"] = sum([enable for (_,enable) in pm.INPUTS_GATE]) * test_value
elif id == 2:
config["NUM_FCN_LAYERS"] = 1
config["NUM_NEURONS_PER_FCN"] = test_value
elif id == 3 or id == 24:
config["NUM_FCN_LAYERS"] = test_value
config["NUM_NEURONS_PER_FCN"] = pm.STATE_DIM[0]*pm.STATE_DIM[1]*2/3
elif id == 4:
config["BUNDLE_ACTION"] = test_value
if test_value == False:
config["ACTION_DIM"] = 2 * pm.SCHED_WINDOW_SIZE + pm.SKIP_TS
elif id == 5:
config["JOB_ARRIVAL_PATTERN"] = test_value
elif id == 6:
config["BATCH_NORMALIZATION"] = test_value
elif id == 7:
config["SL_LOSS_FUNCTION"] = test_value
elif id == 8:
["Norm_Progress", "Job_Progress", "Num_Uncompleted_Jobs"]
if test_value == "Norm_Progress":
config["TS_REWARD_PLUS_JOB_REWARD"] = False
config["NUM_UNCOMPLETED_JOB_REWARD"] = False
elif test_value == "Job_Progress":
config["TS_REWARD_PLUS_JOB_REWARD"] = True
config["NUM_UNCOMPLETED_JOB_REWARD"] = False
elif test_value == "Num_Uncompleted_Jobs":
config["TS_REWARD_PLUS_JOB_REWARD"] = False
config["NUM_UNCOMPLETED_JOB_REWARD"] = True
elif id == 9:
if not test_value:
config["REPLAY_MEMORY_SIZE"] = 256
elif id == 10:
config["VALUE_NET"] = test_value
elif id == 11:
if test_value:
config["INJECT_SAMPLES"] = True
config["EPSILON_GREEDY"] = False
else:
config["INJECT_SAMPLES"] = False
config["EPSILON_GREEDY"] = True
elif id == 12:
config["JOB_ARRIVAL_PATTERN"] = test_value
config["HEURISTIC"] = "DRF"
elif id == 13:
config["JOB_ARRIVAL_PATTERN"] = test_value
config["HEURISTIC"] = "SRTF"
elif id == 14:
config["JOB_ARRIVAL_PATTERN"] = test_value
config["HEURISTIC"] = "Tetris"
elif id == 15:
config["JOB_ARRIVAL_PATTERN"] = test_value
config["HEURISTIC"] = "Optimus"
elif id == 16:
config["HEURISTIC"] = test_value
config["MAX_NUM_WORKERS"] = 8
elif id == 17:
config["NUM_AGENTS"] = test_value
config["MINI_BATCH_SIZE"] = 256/test_value
elif id == 18:
config["CHANGING_JOB_TYPES"] = test_value
elif id == 19:
config["REAL_SPEED_TRACE"] = test_value
elif id == 20:
if test_value == "testbed":
config["TESTBED"] = True
config["CLUSTER_NUM_NODES"] = 6
config["TOT_NUM_JOBS"] = 10
config["MAX_NUM_EPOCHS"] = 1000
config["MAX_ARRVS_PER_TS"] = 5
config["TS_DURATION"] = 300.0
window_size = 4
config["SCHED_WINDOW_SIZE"] = window_size
config["STATE_DIM"] = (sum([enable for (_, enable) in pm.INPUTS_GATE]), window_size)
config["ACTION_DIM"] = 3 * window_size + pm.SKIP_TS
config["NUM_NEURONS_PER_FCN"] = sum([enable for (_, enable) in pm.INPUTS_GATE]) * window_size
elif test_value == "large-1":
config["LARGE_SCALE"] = True
config["CLUSTER_NUM_NODES"] = 100
config["TOT_NUM_JOBS"] = 120
config["MAX_NUM_EPOCHS"] = 80000
config["MAX_ARRVS_PER_TS"] = 6
config["TS_DURATION"] = 1200.0
window_size = 30
config["SCHED_WINDOW_SIZE"] = window_size
config["STATE_DIM"] = (sum([enable for (_, enable) in pm.INPUTS_GATE]), window_size)
config["ACTION_DIM"] = 3 * window_size + pm.SKIP_TS
config["NUM_NEURONS_PER_FCN"] = sum([enable for (_, enable) in pm.INPUTS_GATE]) * window_size
elif test_value == "large-2":
config["LARGE_SCALE"] = True
config["CLUSTER_NUM_NODES"] = 100
config["TOT_NUM_JOBS"] = 180
config["MAX_NUM_EPOCHS"] = 80000
config["MAX_ARRVS_PER_TS"] = 9
config["TS_DURATION"] = 1200.0
window_size = 36
config["SCHED_WINDOW_SIZE"] = window_size
config["STATE_DIM"] = (sum([enable for (_, enable) in pm.INPUTS_GATE]), window_size)
config["ACTION_DIM"] = 3 * window_size + pm.SKIP_TS
config["NUM_NEURONS_PER_FCN"] = sum([enable for (_, enable) in pm.INPUTS_GATE]) * window_size
elif test_value == "large-3":
config["LARGE_SCALE"] = True
config["CLUSTER_NUM_NODES"] = 120
config["TOT_NUM_JOBS"] = 180
config["MAX_NUM_EPOCHS"] = 80000
config["MAX_ARRVS_PER_TS"] = 9
config["TS_DURATION"] = 1200.0
window_size = 36
config["SCHED_WINDOW_SIZE"] = window_size
config["STATE_DIM"] = (sum([enable for (_, enable) in pm.INPUTS_GATE]), window_size)
config["ACTION_DIM"] = 3 * window_size + pm.SKIP_TS
config["NUM_NEURONS_PER_FCN"] = sum([enable for (_, enable) in pm.INPUTS_GATE]) * window_size
elif test_value == "large-4":
config["LARGE_SCALE"] = True
config["CLUSTER_NUM_NODES"] = 500
config["TOT_NUM_JOBS"] = 600
config["MAX_NUM_EPOCHS"] = 80000
config["MAX_ARRVS_PER_TS"] = 30
config["TS_DURATION"] = 1200.0
config["MAX_NUM_WORKERS"] = 50
window_size = 180
config["SCHED_WINDOW_SIZE"] = window_size
config["STATE_DIM"] = (sum([enable for (_, enable) in pm.INPUTS_GATE]), window_size)
config["ACTION_DIM"] = 3 * window_size + pm.SKIP_TS
config["NUM_NEURONS_PER_FCN"] = sum([enable for (_, enable) in pm.INPUTS_GATE]) * window_size
elif test_value == "large-5":
config["LARGE_SCALE"] = True
config["CLUSTER_NUM_NODES"] = 500
config["TOT_NUM_JOBS"] = 600
config["MAX_NUM_EPOCHS"] = 80000
config["MAX_ARRVS_PER_TS"] = 30
config["TS_DURATION"] = 1200.0
config["MAX_NUM_WORKERS"] = 100
window_size = 180
config["SCHED_WINDOW_SIZE"] = window_size
config["STATE_DIM"] = (sum([enable for (_, enable) in pm.INPUTS_GATE]), window_size)
config["ACTION_DIM"] = 3 * window_size + pm.SKIP_TS
config["NUM_NEURONS_PER_FCN"] = sum([enable for (_, enable) in pm.INPUTS_GATE]) * window_size
elif test_value == "large-6":
config["LARGE_SCALE"] = True
config["CLUSTER_NUM_NODES"] = 500
config["TOT_NUM_JOBS"] = 600
config["MAX_NUM_EPOCHS"] = 80000
config["MAX_ARRVS_PER_TS"] = 30
config["TS_DURATION"] = 1200.0
config["MAX_NUM_WORKERS"] = 100
config["VALUE_NET"] = False
window_size = 180
config["SCHED_WINDOW_SIZE"] = window_size
config["STATE_DIM"] = (sum([enable for (_, enable) in pm.INPUTS_GATE]), window_size)
config["ACTION_DIM"] = 3 * window_size + pm.SKIP_TS
config["NUM_NEURONS_PER_FCN"] = sum([enable for (_, enable) in pm.INPUTS_GATE]) * window_size
elif test_value == "small": # by default
config["CLUSTER_NUM_NODES"] = 48
config["TOT_NUM_JOBS"] = 60
config["MAX_NUM_EPOCHS"] = 80000
config["MAX_ARRVS_PER_TS"] = 3
config["TS_DURATION"] = 1200.0
window_size = 20
config["SCHED_WINDOW_SIZE"] = window_size
config["STATE_DIM"] = (sum([enable for (_, enable) in pm.INPUTS_GATE]), window_size)
config["ACTION_DIM"] = 3 * window_size + pm.SKIP_TS
config["NUM_NEURONS_PER_FCN"] = sum([enable for (_, enable) in pm.INPUTS_GATE]) * window_size
elif id == 21:
config["JOB_RESR_BALANCE"] = test_value
elif id == 22:
if not test_value:
config["POLICY_NN_MODEL"] = None
elif id == 23:
config["JOB_EPOCH_EST_ERROR"] = test_value
elif id == 25:
config["TRAIN_SPEED_ERROR"] = test_value
return config
def process_results(root_dir, exp_name, test_values):
results = dict()
for test_value in test_values:
jcts = []
makespans = []
rewards = []
for j in range(NUM_TEST):
dir = root_dir + exp_name + "_" + str(test_value) + "/" + str(j) + '/'
file = dir+exp_name+"_"+str(test_value)+"/rl_validation.txt"
assert os.path.exists(file)
f = open(file, 'r')
temp_jcts = []
temp_makespans = []
temp_rewards = []
for line in f:
segs = line.replace("\n",'').split(" ")
temp_jcts.append(float(segs[2]))
temp_makespans.append(float(segs[3]))
temp_rewards.append(float(segs[4]))
# find the min jct
min_index = np.argmin(temp_jcts)
jcts.append(temp_jcts[min_index])
makespans.append(temp_makespans[min_index])
rewards.append(temp_rewards[min_index])
results[test_value] = (str(np.average(jcts))+"+-"+str(np.std(jcts)),\
str(np.average(makespans))+"+-"+str(np.std(makespans)),\
str(np.average(rewards))+"+-"+str(np.std(rewards)))
f = open(root_dir+"results.txt", "w")
for item in results.items():
f.write(str(item) + "\n")
f.close()
print results
return results
def _sl_rl(dir, config, device):
# SL
sl_config = copy.deepcopy(sl_config_dict)
for key, value in config.items():
if key not in sl_config: # sl_config_dict has higher priority
sl_config[key] = value
os.system("mkdir -p " + dir)
os.system("cp *.py *.txt " + dir)
replace_params(sl_config, dir)
if TASK_ID != 17:
os.system("cd " + dir + " && CUDA_VISIBLE_DEVICES=" + str(device) + " python train.py")
else:
os.system("cd " + dir + " && python train.py")
time.sleep(3)
# RL
replace_params(config, dir)
if TASK_ID != 17:
os.system("cd " + dir + " && CUDA_VISIBLE_DEVICES=" + str(device) + " python train.py")
else:
os.system("cd " + dir + " && python train.py")
def _baseline(dir, config):
os.system("mkdir -p " + dir)
os.system("cp *.py *.txt " + dir)
replace_params(config, dir)
os.system("cd " + dir + " && python comparison.py")
def run(id, exp_name, test_values):
print "running experiments for", exp_name
tic = time.time()
root_dir = exp_name + "-" + datetime.datetime.today().strftime('%Y%m%d_%H%M%S') + "/"
pool = multiprocessing.Pool(processes=PARALLELISM)
for i in range(len(test_values)):
test_value = test_values[i]
print "testing", exp_name, "with value", test_value
parent_dir = root_dir + exp_name + "_" + str(test_value) + "/"
for j in range(NUM_TEST):
print "round", j
dir = parent_dir + str(j) + '/'
config = get_config(id, exp_name, test_value)
device = (i*NUM_TEST+j)%2
if id in [12, 13, 14, 15]:
# _baseline(dir, config)
pool.apply_async(_baseline, args=(dir, config))
else:
# _sl_rl(dir, config, device)
pool.apply_async(_sl_rl, args=(dir, config, device))
if id in [12, 13, 14, 15]:
time.sleep(0.3)
else:
time.sleep(3)
pool.close()
pool.join()
results = process_results(root_dir, exp_name, test_values)
print "finish testing all values of", exp_name
print "the result is:", results
toc = time.time()
print "elapsed time: ", toc - tic, "seconds"
def main(id):
global PARALLELISM, TASK_ID
TASK_ID = id
if id == 1:
exp_name = "sched_window_size"
test_values = [10, 20, 30, 40, 50, 60]
elif id == 2:
exp_name = "number_of_neurons"
test_values = [16, 32, 64, 96, 128, 160, 192, 256]
elif id == 3:
PARALLELISM = 5
exp_name = "number_of_hidden_layers"
test_values = [1, 2, 3, 4]
elif id == 4:
exp_name = "bundle_action" # bundle false error
test_values = [False, True]
elif id == 5:
exp_name = "job_arrival_distribution"
test_values = ["Ali_Trace", "Uniform", "Poisson", "Google_Trace"]
elif id == 6:
exp_name = "batch_normalization"
test_values = [False, True]
elif id == 7:
exp_name = "sl_loss_function"
test_values = ["Mean_Square", "Cross_Entropy", "Absolute_Difference"]
elif id == 8:
exp_name = "job_reward_function"
test_values = ["Norm_Progress", "Job_Progress", "Num_Uncompleted_Jobs"]
elif id == 9:
exp_name = "experience_replay"
test_values = [False, True]
elif id == 10:
exp_name = "critic_network"
test_values = [False, True]
elif id == 11:
exp_name = "exploration"
test_values = [False, True]
elif id == 12:
exp_name = "DRF_baseline"
test_values = ["Ali_Trace", "Uniform", "Poisson", "Google_Trace"]
elif id == 13:
exp_name = "SRTF_baseline"
test_values = ["Ali_Trace", "Uniform", "Poisson", "Google_Trace"]
elif id == 14:
exp_name = "Tetris_baseline"
test_values = ["Ali_Trace", "Uniform", "Poisson", "Google_Trace"]
elif id == 15:
exp_name = "Optimus_baseline"
test_values = ["Ali_Trace", "Uniform", "Poisson", "Google_Trace"]
elif id == 16:
exp_name = "SL_heuristics"
test_values = ["Optimus", "FIFO", "SRTF"]
elif id == 17:
PARALLELISM = 5
exp_name = "a3c"
test_values = [5, 4, 3, 2, 1]
elif id == 18:
exp_name = "changing_job_types"
test_values = [True]
elif id == 19:
exp_name = "analytical_model"
test_values = [False]
elif id == 20:
exp_name = "cluster_scale"
test_values = ["large-4", "large-5", "large-6", "large-1", "large-2", "large-3", "testbed", "small"]
elif id == 21:
exp_name = "job_resr_balance"
test_values = [True, False]
elif id == 22:
exp_name = "enable_SL_or_not"
test_values = [True, False]
elif id == 23:
exp_name = "estimation_error_num_epoch" # error
test_values = [0.05, 0.1, 0.15, 0.2, 0.25]
elif id == 24:
PARALLELISM = 3
exp_name = "number_of_hidden_layers"
test_values = [5, 6, 7]
elif id == 25:
exp_name = "train_speed_error"
test_values = [0.05, 0.1, 0.15, 0.2, 0.25, 0.3]
run(id, exp_name, test_values)
if __name__ == "__main__":
if len(sys.argv) != 2:
print "a script for running experiment"
print "Usage: please input one of following experiment IDs"
print "1: scheduling window size"
print "2: number of neurons"
print "3: number of hidden layers"
print "4: bundle action"
print "5: job arrival distribution"
print "6: batch normalization"
print "7: sl loss function"
print "8: job reward function"
print "9: experience replay"
print "10: critic network"
print "11: exploration"
print "12: DRF baseline"
print "13: SRTF baseline"
print "14: Tetris baseline"
print "15: Optimus baseline"
print "16: SL heuristics"
print "17: a3c, change train_a3c.py to train.py, change parallelism, make sure a correct total batch size before running"
print "18: changing job types during training"
print "19: training on analytical model"
print "20: cluster scale"
print "21: job resource balance"
print "22: enable SL or not"
print "23: estimation error of epoch number"
print "25: train speed error"
exit(1)
main(int(sys.argv[1]))