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
|
#!/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 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'),
]
defaults = {
# output options
'format' : 'pdf',
'save_script' : False,
'prefix' : '',
# data processing
'cycles' : 2128, # per usec
'extent' : 3,
# formatting options
'binsize' : 0.25,
}
def get_data(fname, scale, extend):
data = load_binary_file(fname)
if not scale is None:
data *= scale
data.sort()
iqr_min, iqr_max = iqr_cutoff(data, extend)
min_idx, max_idx = numpy.searchsorted(data, [iqr_min, iqr_max])
return [data, max_idx, min_idx, iqr_max, iqr_min]
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))
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
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'))
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)
samples = data[min_idx:max_idx]
discarded = (len(data) - len(samples)) / float(len(data)) * 100
max_cost = data[-1]
p = self.make_plot(name)
iqr_label = "IQR: extent=%d threshold=%.2fus filtered=%.2f%%" % \
(self.options.extent, iqr_max, discarded)
p.labels = [label(0.5, 0.9,
get_stats_label(samples),
coord=['graph', 'screen'], align='center'),
label(0.98, 0.95, iqr_label,
coord=['graph', 'graph'], align='right')]
(hist, fname) = self.write_histogram(samples, name)
p.setup_histogram(gap=1, boxwidth=1.0)
p.title = "measured overheads scheduler=%s; overhead=%s; host=%s" \
% (conf['scheduler'], conf['overhead'], conf['host'])
p.ylabel = "number of samples"
p.xlabel = "overhead in microseconds (bin size = %.2fus)" \
% self.options.binsize
# p.xrange = (0, ceil(max_cost))
p.xticks = (0, 10)
# p.yticks = (0, 1)
p.yrange = (0, (ceil(numpy.amax(hist[:,1]) / 100.0) * 100))
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 plot_file(self, datafile):
bname = basename(datafile)
name, ext = splitext(bname)
conf = decode(name)
plotters = {
'taskset' : self.plot_samples,
}
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()
|
