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
|
#!/usr/bin/env python
import defapp
from plot import decode
from util import load_csv_file, load_binary_file, write_csv_file
from stats import iqr_cutoff
from binary_data import get_data
from math import ceil
import numpy
from os.path import splitext, basename
from optparse import make_option as o
from gnuplot import gnuplot, FORMATS, Plot, label, curve
options = [
# output options
o('-f', '--format', action='store', dest='format', type='choice',
choices=FORMATS, help='output format'),
o(None, '--save-script', action='store_true', dest='save_script'),
o('-p', '--prefix', action='store', dest='prefix'),
o('-i', '--iqr-extent', action='store', dest='extent', type='float',
help='what extent to use for outlier removal'),
o('-n', '--normalize', action='store_true', dest='normalize',
help='use normalize counts'),
o('-c', '--cut-off', action='store', dest='cutoff', type='int',
help='max number of samples to use'),
o('-x', '--xmax', action='store', dest='xmax', type='int',
help='determines x-axis range'),
o('-y', '--ymax', action='store', dest='ymax', type='float',
help='determines y-axis range'),
o('-b', '--binsize', action='store', dest='binsize', type='float',
help='set binsize of histogram'),
]
defaults = {
# output options
'format' : 'pdf',
'save_script' : False,
'prefix' : '',
# data processing
'cycles' : 2128, # per usec
'extent' : 3,
'cutoff' : None,
'normalize' : False,
# formatting options
'binsize' : 0.25,
'xmax' : None,
'ymax' : None,
}
TXT = {
'RELEASE-LATENCY' : 'timer latency',
'RELEASE' : 'job release overhead',
'SCHED' : 'scheduling overhead',
'SCHED2' : 'post-scheduling overhead',
'CXS' : 'context-switch overhead',
'SEND-RESCHED' : 'IPI latency',
'TICK' : 'timer tick overhead',
}
HOST_CPUS = {
'ludwig' : 24,
}
def get_stats_label(samples):
avg = numpy.mean(samples)
med = numpy.median(samples)
dev = numpy.std(samples)
max = samples[-1]
min = samples[0]
return "min=%.2fus max=%.2fus avg=%.2fus median=%.2fus stdev=%.2fus" \
% (min, max, avg, med, dev)
class OverheadPlotter(defapp.App):
def __init__(self):
defapp.App.__init__(self, options, defaults, no_std_opts=True)
self.tmpfiles = []
def make_plot(self, fname=None):
p = Plot()
p.output = "%s%s.%s" % (self.options.prefix, fname, self.options.format)
p.format = self.options.format
return p
def setup_png(self, plot):
# standard png options; usually correct; never tweaked for paper
if self.options.format == 'png':
plot.font_size = 'large'
plot.size = (1024, 768)
plot.xticks = (0, 1)
plot.yticks = (0, 0.1)
plot.default_style = "linespoints"
return True
else:
return False
def write(self, data, name, ext='data'):
if self.options.save_script:
fname = "%s.%s" % (name, ext)
write_csv_file(fname, data)
return fname
else:
tmp = write_csv_file(None, data)
# keep a reference so that it isn't deleted
self.tmpfiles.append(tmp)
return tmp.name
def write_histogram(self, samples, name, labels=10):
max = ceil(numpy.amax(samples))
if self.options.xmax:
max = self.options.xmax
bin_size = self.options.binsize
num_bins = int(max / bin_size)
(bins, edges) = numpy.histogram(samples, bins=num_bins,
range=(self.options.binsize / 2,
max + self.options.binsize / 2))
data = numpy.zeros((num_bins, 3))
cumulative = 0
for i in xrange(len(bins)):
data[i, 0] = (edges[i] + edges[i + 1]) / 2.0
data[i, 1] = bins[i]
cumulative += bins[i]
data[i, 2] = cumulative
if self.options.normalize:
data[:, 1] /= len(samples)
data[:, 2] /= len(samples)
label_rate = len(bins) / labels
if not label_rate:
label_rate = 1
for_file = []
for i, row in enumerate(data):
label = '%.2f' % row[0] if i % label_rate == 0 else ''
for_file.append([row[0], row[1], row[2], label])
return (data, self.write(for_file, name, ext='hist'), edges)
def render(self, p):
if self.options.save_script:
p.gnuplot_save(p.output + '.plot')
else:
p.gnuplot_exec()
def plot_samples(self, datafile, name, conf):
if conf['overhead'] == 'RELEASE-LATENCY':
scale = 1.0 / 1000.0
else:
scale = 1.0 / self.options.cycles
data, max_idx, min_idx, iqr_max, iqr_min = get_data(datafile,
scale,
self.options.extent,
self.options.cutoff)
samples = data[min_idx:max_idx]
discarded = (len(data) - len(samples)) / float(len(data)) * 100
max_cost = data[-1]
p = self.make_plot(name)
samples_label = "samples: total=%d filtered=%d (%.2f%%)" % \
(len(data), len(data) - len(samples), discarded)
if self.options.extent:
iqr_label = "IQR: extent=%d threshold=%.2fus" % \
(self.options.extent, iqr_max)
elif discarded > 0:
iqr_label = "manual threshold=1000us [IQR not applied]"
else:
iqr_label = "[IQR filter not applied]"
samples_label = "samples: total=%d" % len(data)
data_label = "%s\\n%s" % (samples_label, iqr_label)
p.labels = [label(0.5, 0.9,
get_stats_label(samples),
coord=['graph', 'screen'], align='center'),
label(0.98, 0.95, data_label,
coord=['graph', 'graph'], align='right')]
(hist, fname, edges) = self.write_histogram(samples, name)
p.setup_histogram(gap=1, boxwidth=1.0)
p.title = "%s: measured %s for %s task%s per processor (host=%s)" \
% (conf['scheduler'], TXT[conf['overhead']],
conf['n'], 's' if conf['n'] != '1' else '',
conf['host'])
if self.options.normalize:
p.ylabel = "fraction of samples"
else:
p.ylabel = "number of samples"
p.xlabel = "overhead in microseconds (bin size = %.2fus)" \
% self.options.binsize
if self.options.ymax:
p.yrange = (0, self.options.ymax)
# p.yrange = (0, (ceil(numpy.amax(hist[:,1]) / 100.0) * 100))
p.xticks = (0, 10)
p.curves = [curve(histogram=fname, col=2, labels_col=4)]
#### Styling.
if not self.setup_png(p):
p.rounded_caps = True
p.font = 'Helvetica'
p.font_size = '10'
p.size = ('20cm', '10cm')
p.monochrome = False
p.dashed_lines = False
p.key = 'off'
p.default_style = 'points lw 1'
self.render(p)
def prepare_trends(self, datafile, name, conf):
data = load_csv_file(datafile)
if 'host' in conf and conf['host'] in HOST_CPUS:
cpus = HOST_CPUS[conf['host']]
if conf['scheduler'].endswith('-RM'):
cpus -= 1
else:
cpus = 1
# format
n_idx = 2
wc_idx = 5
avg_idx = 6
std_idx = 9
rows = [ [r[n_idx] * cpus,
r[wc_idx],
r[avg_idx],
r[std_idx]]
for r in data]
max_y = numpy.amax(data[:,wc_idx])
return (self.write(rows, name), cpus, max_y)
def plot_trends(self, datafile, name, conf):
fname, cpus, max_y = self.prepare_trends(datafile, name, conf)
p = self.make_plot(name)
p.title = "measured %s under %s scheduling" \
% (TXT[conf['overhead']], conf['scheduler'])
p.ylabel = "overhead in microseconds"
p.xlabel = "number of tasks"
if self.options.xmax:
p.xrange = (0, self.options.xmax)
else:
p.xrange = (0, ceil(cpus * 20 / 100.0) * 100)
if self.options.ymax:
p.yrange = (0, self.options.ymax)
else:
p.yrange = (0, (ceil(max_y / 50.0)) * 50)
p.xticks = (0, max(cpus, 10))
p.curves = [curve(fname, xcol=1, ycol=2, title="maximum"),
curve(fname, xcol=1, ycol=3, style="lines", title='average'),
curve(fname, xcol=1, ycol=3, error=4, title="std. deviation")]
#### Styling.
if not self.setup_png(p):
p.rounded_caps = True
p.font = 'Helvetica'
p.font_size = '10'
p.size = ('20cm', '10cm')
p.monochrome = False
p.dashed_lines = False
p.key = 'left top'
p.default_style = 'linespoints lw 1'
self.render(p)
def plot_file(self, datafile):
bname = basename(datafile)
name, ext = splitext(bname)
conf = decode(name)
plotters = {
'taskset' : self.plot_samples,
'otrend' : self.plot_trends,
}
for plot_type in plotters:
if plot_type in conf:
try:
plotters[plot_type](datafile, name, conf)
except IOError as err:
self.err("Skipped '%s' (%s)." % err)
break
else:
self.err("Skipped '%s'; unkown experiment type."
% bname)
# release all tmp files
self.tmpfiles = []
def default(self, _):
for i, datafile in enumerate(self.args):
self.out("[%d/%d] Processing %s ..." % (i + 1, len(self.args), datafile))
self.plot_file(datafile)
if __name__ == "__main__":
OverheadPlotter().launch()
|
