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
441
442
443
444
445
446
447
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
484
485
486
487
488
489
490
491
492
content / test / gpu / power_measurement_results / analyze_power_measurement_results.py [blame]
#!/usr/bin/env python
# Copyright 2019 The Chromium Authors
# Use of this source code is governed by a BSD-style license that can be
# found in the LICENSE file.
"""Script which analyze power measurement test results from bots.
Such analysis provides insights into power data to better understand
Intel Power Gadget.
Related design doc:
https://docs.google.com/document/d/1s3L2IYguQmPHInsKkbHh06hXCXo8ggo5iPIhOaCNwVw
"""
import enum
import json
import logging
import math
import os
import sys
_TESTS = [
'Basic', 'Video_720_MP4', 'Video_720_MP4_Fullscreen',
'Video_720_MP4_Underlay', 'Video_720_MP4_Underlay_Fullscreen'
]
_MEASUREMENTS = ['DRAM', 'Processor']
_RESULTS_PATH = os.path.join(
os.path.dirname(os.path.abspath(__file__)), 'win10_intel_hd_630')
_RESULTS_JSON_FILES = [
'build_4370_4425_repeat3.json',
'build_4426_4759_repeat3.json',
'build_4760_5047_repeat3.json',
]
MIN_RUNS_PER_BOT = 8
class RepeatStrategy(enum.Enum):
COUNT_EACH = 1 # count each run individually
COUNT_MINIMUM = 2 # count the run with minimum power
COUNT_AVERAGE = 3 # count the average power of all runs
COUNT_MEDIAN = 4 # count the median_low (power wise) of all runs
COUNT_MINIMUM_FIRST_TWO = 5 # for the first 2, count the run with less power
@classmethod
def ToString(cls, strategy):
if strategy == cls.COUNT_EACH:
return 'each'
if strategy == cls.COUNT_MINIMUM:
return 'minimum'
if strategy == cls.COUNT_AVERAGE:
return 'average'
if strategy == cls.COUNT_MEDIAN:
return 'median'
assert strategy == cls.COUNT_MINIMUM_FIRST_TWO
return 'minimum (first two)'
def LoadResultsJsonFiles():
jsons = []
for json_filename in _RESULTS_JSON_FILES:
json_path = os.path.join(_RESULTS_PATH, json_filename)
with open(json_path, 'r') as json_file:
logging.debug('Loading %s', json_path)
jsons.append(json.load(json_file))
return jsons
def DetermineResultsFromMultipleRuns(measurements, repeat_strategy):
if repeat_strategy == RepeatStrategy.COUNT_EACH:
return measurements
if repeat_strategy == RepeatStrategy.COUNT_MINIMUM:
measurements.sort()
return [measurements[0]]
if repeat_strategy == RepeatStrategy.COUNT_AVERAGE:
return [Mean(measurements)]
if repeat_strategy == RepeatStrategy.COUNT_MEDIAN:
return [MedianLow(measurements)]
if repeat_strategy == RepeatStrategy.COUNT_MINIMUM_FIRST_TWO:
assert len(measurements) >= 2
first_two = measurements[0:2]
first_two.sort()
return [first_two[0]]
assert False
return []
# pylint: disable=too-many-locals
def ProcessJsonData(jsons,
per_bot=False,
repeat_strategy=RepeatStrategy.COUNT_MINIMUM):
min_build = sys.maxsize
max_build = -1
results = {}
bots = set()
for j in jsons:
for build in j.get('builds', []):
build_number = build.get('number', -1)
if build_number > 0:
min_build = min(build_number, min_build)
max_build = max(build_number, max_build)
bot = build['bot']
bots.add(bot)
for test in build['tests']:
name = test['name'].split('.')[-1]
assert name in _TESTS
if results.get(name, None) is None:
if per_bot:
results[name] = {}
else:
results[name] = []
test_data = results[name]
if per_bot:
if test_data.get(bot, None) is None:
test_data[bot] = []
test_data = test_data[bot]
measurements = [0]
for measurement_name in _MEASUREMENTS:
actual_measurement_name = measurement_name + ' Power_0'
data = test[actual_measurement_name]
count = len(data)
while len(measurements) < count:
measurements.append(0)
for ii in range(count):
measurements[ii] = measurements[ii] + data[ii]
assert measurements
test_data.extend(
DetermineResultsFromMultipleRuns(measurements, repeat_strategy))
return {
'min_build': min_build,
'max_build': max_build,
'bots': list(bots),
'results': results,
}
# pylint: enable=too-many-locals
def Mean(data):
assert len(data) > 0
total = 0
for num in data:
total = total + num
return total / len(data)
def Stdev(data):
assert len(data) > 0
mean = Mean(data)
total = 0
for num in data:
total = total + (num - mean) * (num - mean)
return math.sqrt(total / len(data))
def MedianLow(data):
# Assume list is sorted.
assert len(data) > 0
index = int((len(data) - 1) / 2)
return data[index]
def MarkSection():
print('')
def MarkExperiment(description):
print('')
print('**************************************************************')
print(description)
print('**************************************************************')
print('')
def GetBotBuilds(jsons, bot_name):
build_numbers = []
for j in jsons:
builds = j.get('builds', [])
for build in builds:
build_number = build.get('number', -1)
if build_number > 0:
bot = build['bot']
if bot == bot_name:
build_numbers.append(build_number)
return build_numbers
def GetOutliers(data, variation_threshold):
mean = Mean(data)
max_value = mean + mean * variation_threshold
min_value = mean - mean * variation_threshold
outliers = []
for value in data:
if value > max_value or value < min_value:
outliers.append(value)
return outliers
# pylint: disable=too-many-locals
def FindBuild(jsons, selected_bots, test_name, result):
for j in jsons:
for build in j.get('builds', []):
build_number = build.get('number', -1)
if build_number < 0:
continue
bot = build['bot']
if bot not in selected_bots:
continue
for test in build['tests']:
name = test['name'].split('.')[-1]
assert name in _TESTS
if name != test_name:
continue
# Use RepeatStrategy.COUNT_MINIMUM
measurements = [0]
for measurement_name in _MEASUREMENTS:
actual_measurement_name = measurement_name + ' Power_0'
data = test[actual_measurement_name]
count = len(data)
while len(measurements) < count:
measurements.append(0)
for ii in range(count):
measurements[ii] = measurements[ii] + data[ii]
assert measurements
measurements.sort()
if measurements[0] == result:
return {'bot': bot, 'build': build_number}
return None
# pylint: enable=too-many-locals
def RunExperiment_BadBots(jsons,
stdev_threshold,
repeat_strategy=RepeatStrategy.COUNT_MINIMUM):
MarkExperiment('Locate potential bad bots: thresh=%0.2f, repeat=%s' %
(stdev_threshold, RepeatStrategy.ToString(repeat_strategy)))
outcome = ProcessJsonData(
jsons, per_bot=True, repeat_strategy=repeat_strategy)
logging.debug('Processed builds: [%d, %d]', outcome['min_build'],
outcome['max_build'])
logging.debug('Total number of bots: %d', len(outcome['bots']))
results = outcome['results']
total_bad_bots = set()
for test_name, test_results in results.items():
if test_name == 'Basic':
# Ignore Basic test results. They seem more unstable.
continue
MarkSection()
logging.debug('Results for test: %s', test_name)
bots_considered = 0
bad_bots = []
for bot_name, bot_results in test_results.items():
if len(bot_results) < MIN_RUNS_PER_BOT:
continue
bots_considered = bots_considered + 1
stdev = Stdev(bot_results)
bot_results.sort()
if stdev > stdev_threshold:
bad_bots.append(bot_name)
logging.debug('Potential bad bot %s: stdev = %f', bot_name, stdev)
total_bad_bots |= set(bad_bots)
logging.debug('Total bots considered: %d', bots_considered)
logging.debug('Bad bots: %d', len(bad_bots))
logging.debug('%s', bad_bots)
MarkSection()
total_bad_bots = list(total_bad_bots)
total_bad_bots.sort()
logging.debug('All potential bad bots: %d', len(total_bad_bots))
logging.debug('%s', total_bad_bots)
MarkSection()
for bot in total_bad_bots:
build_numbers = GetBotBuilds(jsons, bot)
build_numbers.sort()
logging.debug('Bad bot %s builds: %s', bot, build_numbers)
return total_bad_bots
# pylint: disable=too-many-locals
def RunExperiment_GoodBots(jsons,
bad_bots=None,
repeat_strategy=RepeatStrategy.COUNT_MINIMUM):
bad_bots = bad_bots or []
STDEV_GOOD_BOT_THRESHOLD = 0.2
GOOD_BOT_RANGE_PERC = 0.08
REGULAR_BOT_RANGE_PERC = 0.15
MarkExperiment(
'Locate potential good bots: thresh=%0.2f, repeat=%s' %
(STDEV_GOOD_BOT_THRESHOLD, RepeatStrategy.ToString(repeat_strategy)))
outcome = ProcessJsonData(
jsons, per_bot=True, repeat_strategy=repeat_strategy)
logging.debug('Processed builds: [%d, %d]', outcome['min_build'],
outcome['max_build'])
logging.debug('Total number of bots: %d', len(outcome['bots']))
total_good_bots = set(outcome['bots'])
for test_name, test_results in outcome['results'].items():
if test_name == 'Basic':
# Ignore Basic test results. They seem more unstable.
continue
MarkSection()
logging.debug('Results for test: %s', test_name)
bots_considered = 0
stdev_list = []
good_bots = []
for bot_name, bot_results in test_results.items():
if len(bot_results) < MIN_RUNS_PER_BOT:
continue
stdev = Stdev(bot_results)
stdev_list.append(stdev)
if bot_name in bad_bots:
continue
bot_results.sort()
bots_considered = bots_considered + 1
mean = Mean(bot_results)
if stdev < STDEV_GOOD_BOT_THRESHOLD:
good_bots.append(bot_name)
logging.debug('Potential good bot %s: mean = %f, stdev = %f', bot_name,
mean, stdev)
outliers = GetOutliers(bot_results, GOOD_BOT_RANGE_PERC)
if outliers:
logging.debug('Good bot %s: %d runs out of %d%% range', bot_name,
len(outliers), GOOD_BOT_RANGE_PERC * 100)
else:
outliers = GetOutliers(bot_results, REGULAR_BOT_RANGE_PERC)
if outliers:
logging.debug('Regular bot %s: %d runs out of %d%% range', bot_name,
len(outliers), REGULAR_BOT_RANGE_PERC * 100)
total_good_bots &= set(good_bots)
logging.debug('Total bots considered: %d', bots_considered)
logging.debug('Good bots: %d', len(good_bots))
logging.debug('%s', good_bots)
logging.debug('Average per bot stdev: %f', Mean(stdev_list))
MarkSection()
total_good_bots = list(total_good_bots)
total_good_bots.sort()
logging.debug('All potential good bots: %d', len(total_good_bots))
logging.debug('%s', total_good_bots)
MarkSection()
for bot in total_good_bots:
build_numbers = GetBotBuilds(jsons, bot)
build_numbers.sort()
logging.debug('Good bot %s builds: %s', bot, build_numbers)
return total_good_bots
# pylint: enable=too-many-locals
# This could definitely use some refactoring to be more readable and make
# pylint happier, but currently difficult to change confidently without any
# unittests.
# pylint: disable=too-many-locals,too-many-branches,too-many-statements
def RunExperiment_BestVariations(jsons, find_m_bots, variation_threshold):
GET_RID_OF_N_BOTS_WITH_WORST_STDEV = 10
MarkExperiment('Find %d bots with best variations, threshold = %0.2f%%' %
(find_m_bots, (variation_threshold * 100)))
outcome = ProcessJsonData(
jsons, per_bot=True, repeat_strategy=RepeatStrategy.COUNT_MINIMUM)
candidates_per_test = {}
candidate_bots_per_test = {}
for test_name, test_results in outcome['results'].items():
if test_name == 'Basic':
# Ignore Basic test results. They seem more unstable.
continue
bots_considered = 0
candidates = []
stdev_list = []
# Remove N bots with worst stdev
for bot_name, bot_results in test_results.items():
if len(bot_results) < MIN_RUNS_PER_BOT:
continue
bots_considered = bots_considered + 1
mean = Mean(bot_results)
stdev = Stdev(bot_results)
candidates.append({
'bot': bot_name,
'mean': mean,
'stdev': stdev,
'data': bot_results,
})
stdev_list.append(stdev)
stdev_list.sort()
guard_stdev = stdev_list[-GET_RID_OF_N_BOTS_WITH_WORST_STDEV]
candidates_with_good_stdev = []
mean_list = []
for candidate in candidates:
if candidate['stdev'] < guard_stdev:
candidates_with_good_stdev.append(candidate)
mean_list.append(candidate['mean'])
assert (len(candidates) - GET_RID_OF_N_BOTS_WITH_WORST_STDEV == len(
candidates_with_good_stdev))
assert len(candidates_with_good_stdev) > find_m_bots
# Find M bots with minimum range of means
mean_list.sort()
min_range = mean_list[-1] - mean_list[0]
candidate_index = 0
for low_index in range(len(candidates_with_good_stdev) - find_m_bots + 1):
high_index = low_index + find_m_bots - 1
mean_range = mean_list[high_index] - mean_list[low_index]
if mean_range < min_range:
min_range = mean_range
candidate_index = low_index
min_mean = mean_list[candidate_index]
max_mean = mean_list[candidate_index + find_m_bots - 1]
candidates = []
candidate_bots = []
for candidate in candidates_with_good_stdev:
if candidate['mean'] >= min_mean and candidate['mean'] <= max_mean:
candidates.append(candidate)
candidate_bots.append(candidate['bot'])
assert len(candidates) == find_m_bots
candidate_bots_per_test[test_name] = set(candidate_bots)
candidates_per_test[test_name] = candidates
# Now find the list of bots that work well for all tests
selected_bots = None
for test_name, bots in candidate_bots_per_test.items():
if selected_bots is None:
selected_bots = bots
else:
selected_bots = selected_bots & bots
logging.debug('Intended to find %d bots, actually found %d', find_m_bots,
len(selected_bots))
selected_bots = list(selected_bots)
selected_bots.sort()
logging.debug(selected_bots)
# Validate: check variations are within a range
for test_name, candidates in candidates_per_test.items():
MarkSection()
results = []
for candidate in candidates:
if candidate['bot'] in selected_bots:
results.extend(candidate['data'])
mean = Mean(results)
stdev = Stdev(results)
logging.debug('Validate test %s: mean = %f, stdev = %f', test_name, mean,
stdev)
outliers = GetOutliers(results, variation_threshold)
if outliers:
# Find corresponding builds
builds = []
for outlier in outliers:
build = FindBuild(jsons, selected_bots, test_name, outlier)
assert build is not None
builds.append(build)
logging.debug('%d runs out of %d are not within %0.2f%% range: %s',
len(outliers), len(results), (variation_threshold * 100),
outliers)
logging.debug(builds)
# pylint: enable=too-many-locals,too-many-branches,too-many-statements
def main():
logging.basicConfig(level=logging.DEBUG)
jsons = LoadResultsJsonFiles()
bad_bots = RunExperiment_BadBots(jsons, 0.5, RepeatStrategy.COUNT_EACH)
RunExperiment_GoodBots(jsons, bad_bots, RepeatStrategy.COUNT_EACH)
bad_bots = RunExperiment_BadBots(jsons, 0.5, RepeatStrategy.COUNT_AVERAGE)
RunExperiment_GoodBots(jsons, bad_bots, RepeatStrategy.COUNT_AVERAGE)
bad_bots = RunExperiment_BadBots(jsons, 0.5, RepeatStrategy.COUNT_MEDIAN)
RunExperiment_GoodBots(jsons, bad_bots, RepeatStrategy.COUNT_MEDIAN)
bad_bots = RunExperiment_BadBots(jsons, 0.5,
RepeatStrategy.COUNT_MINIMUM_FIRST_TWO)
RunExperiment_GoodBots(jsons, bad_bots,
RepeatStrategy.COUNT_MINIMUM_FIRST_TWO)
bad_bots = RunExperiment_BadBots(jsons, 0.5, RepeatStrategy.COUNT_MINIMUM)
RunExperiment_GoodBots(jsons, bad_bots, RepeatStrategy.COUNT_MINIMUM)
RunExperiment_BestVariations(jsons, 25, 0.12)
return 0
if __name__ == '__main__':
sys.exit(main())