-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathct_experiment.py
311 lines (253 loc) · 11.3 KB
/
ct_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
"""
This file defines the top level functions:
(1) setupExperiment
(2) runExperiment
(3) reportResults
"""
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
from matplotlib.backends.backend_pdf import PdfPages
import numpy as np
import subprocess, shlex, math, sys, os, socket, time
from workload import Workload
from mp_max_min import MPMaxMin
from get_ctime import *
from tcpprobe import *
#BUFSIZE = 2048
SND_MSG_LEN = 1500*10000
SND_BUFSIZE = 2**21 #2048 #4096 #262144
RCV_LEN = 4096
DURATION = 10
CUTOFF = 10
TCP_VERSION = "reno"
class CT_Experiment:
"""
starts the rate monitor on each host
starts tcp server on each destination machine
"""
def __init__(self, workload):
self.workload = workload
self.logging_processes = []
self.tcp_servers = []
self.tcp_clients = []
kill_logging = 'ssh root@{0} "pkill -u root cat"'
unload_tcp_probe = 'ssh root@{0} "modprobe -r tcp_probe"'
load_tcp_probe = 'ssh root@{0} "modprobe tcp_probe port=0 full=1"' # port=0 means match on all ports
log_file = '/tmp/tcpprobe_{0}.log'
write_log_file = 'ssh root@{0} "cat /proc/net/tcpprobe >%s"' % log_file
set_tcp_version = 'ssh root@{0} "echo {1} > /proc/sys/net/ipv4/tcp_congestion_control"'
for host in workload.allHosts:
# kill all currently running logging processes if there are any
command = kill_logging.format(host)
rc = self.runCommand(command)
if rc not in [0,1]:
print >> sys.stderr, "ERROR: {0} -- failed".format(command)
# load tcp_probe on the source host
self.runCommand(unload_tcp_probe.format(host))
self.runCommand(load_tcp_probe.format(host))
p = self.startProcess(write_log_file.format(host))
self.logging_processes.append((host, p))
# set TCP congestion control algorithm to use
self.runCommand(set_tcp_version.format(host, TCP_VERSION))
for flow in workload.flows:
self.setupFlow(flow)
# sleep to give the servers time to set up
time.sleep(3)
def setupFlow(self, flow):
start_tcp_server = os.path.expandvars('ssh root@{0} "$CT_EXP_DIR/run_server.py {1} {2} {3}"'.format(flow['dstHost'], flow['dstIP'], flow['port'], RCV_LEN))
# start tcp server on the destination
p = self.startProcess(start_tcp_server)
self.tcp_servers.append((flow['dstHost'], p))
"""
get the global start time, distributed to each of the
required hosts, and run the experiment
"""
def runExperiment(self):
currTime = get_real_time()
expStartTime = int(math.floor(currTime + 5)) # start the experiment 5 seconds from now
start_tcp_client = os.path.expandvars('ssh root@{0} "$CT_EXP_DIR/run_client.py {1} {2} {3} {4} {5} {6}"')
# start tcp clients on each src machine
for flow in self.workload.flows:
command = start_tcp_client.format(flow['srcHost'], flow['dstIP'], flow['port'], SND_BUFSIZE, SND_MSG_LEN, expStartTime, DURATION)
p = self.startProcess(command)
self.tcp_clients.append((flow['srcHost'], p))
# wait for all tcp clients to finish
for (host, tcp_client) in self.tcp_clients:
print "Waiting for tcp client on host {0} ...".format(host)
tcp_client.wait()
rc = tcp_client.returncode
print "tcp client on host {0} finished with return code: {1}".format(host, rc)
if rc != 0:
output, error = tcp_client.communicate()
print("tcp_client failed %s %s" % (output, error))
self.cleanupExperiment()
# kill all tcpprobe logging processes
# kill all tcp servers
# copy all tcpprobe log files to single location
def cleanupExperiment(self):
# kill all tcpprobe logging processes
for (host, log_process) in self.logging_processes:
log_process.kill()
command = 'ssh root@{0} "pkill -u root cat"'.format(host)
rc = self.runCommand(command)
if rc not in [0,1]:
print >> sys.stderr, "ERROR: {0} -- failed".format(command)
# kill all tcp servers
for (host, server) in self.tcp_servers:
server.kill()
command = 'ssh root@{0} "pkill -u root python"'.format(host)
rc = self.runCommand(command)
if rc not in [0,1]:
print >> sys.stderr, "ERROR: {0} -- failed".format(command)
log_dir = os.path.expandvars('$CT_EXP_DIR/logs/')
copy_log_file = 'scp root@{0}:~/../tmp/tcpprobe_* %s' % log_dir
# copy all log files to single location
if not os.path.exists(log_dir):
os.makedirs(log_dir)
for host in self.workload.allHosts:
self.runCommand(copy_log_file.format(host))
def runCommand(self, command):
print "----------------------------------------"
print "Running Command:\n"
print "-->$ ", command
print "----------------------------------------"
return subprocess.call(command, shell=True)
def startProcess(self, command):
print "----------------------------------------"
print "Starting Process:\n"
print "-->$ ", command
print "----------------------------------------"
return subprocess.Popen(shlex.split(command), stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
# parse the tcpprobe log files in $CT_EXP_DIR/logs to determine
# convergence time of each flow
def getResults(self):
results = {}
results['rates'] = [] # entry i is like: (time_i, rate_i)
results['convTimes'] = []
results['cwnd'] = [] # entry i is like: (time_i, cwnd_i)
results['srtt'] = [] # entry i is like: (time_i, srtt_i)
# calculate the ideal rates for the particular workload
wf = MPMaxMin(self.workload)
idealRates = wf.maxmin_x
# get the rates and convergence times of each flow
for flowID, flow in zip(range(len(self.workload.flows)), self.workload.flows):
host = self.workload.ipHostMap[flow['srcIP']]
logFile = os.path.expandvars('$CT_EXP_DIR/logs/tcpprobe_{0}.log'.format(host))
time1, rate, _cwnd, _srtt = get_tcpprobe_stats(logFile, flow['srcIP'], flow['dstIP'], flow['port'])
time2, cwnd, srtt = get_tcpprobe_cwnd_srtt(logFile, flow['srcIP'], flow['dstIP'], flow['port'])
results['rates'].append((time1, rate))
results['convTimes'].append(self.getCTtime(time1, rate, idealRates[flowID]))
results['cwnd'].append((time2, cwnd))
results['srtt'].append((time2, srtt))
return results
"""
Determine how long it takes for rate to converge to idealRate.
time: list of time values (sec)
convergence is defined as within 10% for at least 1ms
"""
def getCTtime(self, time, rate, idealRate):
dur = 0.001;
interval = time[1] - time[0]
percent = 0.1
numSampNeeded = math.ceil(dur/interval)
numConverged = 0
for t, r in zip(time, rate):
if r >= (1-percent)*idealRate and r <= (1+percent)*idealRate:
numConverged += 1
if numConverged == numSampNeeded:
return t # converged!
else:
numConverged = 0
return -1 # failed to converge
def reportResults(self, results):
self.out_dir = os.path.expandvars('$CT_EXP_DIR/out')
if not os.path.exists(self.out_dir):
os.makedirs(self.out_dir)
os.system(os.path.expandvars('rm $CT_EXP_DIR/out/*'))
self.plotFlowRates(results)
self.plotCwnd(results)
self.plotSrtt(results)
self.plotCTCDF(results)
def plotCTCDF(self, results):
# plot the CDF of convergence times for each flow
sortData = np.sort(results['convTimes'])
yvals = np.arange(len(sortData))/float(len(sortData))
plt.plot(sortData, yvals, marker='o')
plt.ylabel('CDF')
plt.xlabel('Convergence Time (sec)')
plt.title('CDF of convergence times')
plt.grid()
base_fname = self.out_dir + '/CT_CDF'
self.recordData(sortData, yvals, base_fname + '.csv')
plot_filename = base_fname + '.pdf'
pp = PdfPages(plot_filename)
pp.savefig()
pp.close()
print "Saved plot: ", plot_filename
def plotFlowRates(self, results):
cutoff = CUTOFF
# plot all flow rates on single plot for now
for (flowID, (time, rate)) in zip(range(len(results['rates'])), results['rates']):
csv_file = self.out_dir + '/flow_{0}_rate.csv'.format(flowID)
t, r = self.cutToTime(time, rate, cutoff)
self.recordData(t, r, csv_file)
plt.plot(t, r, label='flow {0}'.format(flowID), marker='o')
plt.legend()
plt.title('Flow Rates over time')
plt.xlabel('time (sec)')
plt.ylabel('rate (Gbps)')
plot_filename = self.out_dir + '/flow_rates.pdf'
pp = PdfPages(plot_filename)
pp.savefig()
pp.close()
print "Saved plot: ", plot_filename
plt.cla()
def plotCwnd(self, results):
cutoff = CUTOFF
# plot all flow rates on single plot for now
for (flowID, (time, cwnd)) in zip(range(len(results['cwnd'])), results['cwnd']):
csv_file = self.out_dir + '/flow_{0}_cwnd.csv'.format(flowID)
t, c = self.cutToTime(time, cwnd, cutoff)
self.recordData(t, c, csv_file)
plt.plot(t, c, label='flow {0}'.format(flowID), marker='o')
plt.legend()
plt.title('Congestion Window over time')
plt.xlabel('time (sec)')
plt.ylabel('congestion window')
plot_filename = self.out_dir + '/flow_cwnd.pdf'
pp = PdfPages(plot_filename)
pp.savefig()
pp.close()
print "Saved plot: ", plot_filename
plt.cla()
def plotSrtt(self, results):
cutoff = CUTOFF
# plot all flow rates on single plot for now
for (flowID, (time, srtt)) in zip(range(len(results['srtt'])), results['srtt']):
csv_file = self.out_dir + '/flow_{0}_srtt.csv'.format(flowID)
t, s = self.cutToTime(time, srtt, cutoff)
self.recordData(t, s, csv_file)
plt.plot(t, s, label='flow {0}'.format(flowID), marker='o')
plt.legend()
plt.title('Smoothed RTT over time')
plt.xlabel('time (sec)')
plt.ylabel('Smoothed RTT (us)')
plot_filename = self.out_dir + '/flow_srtt.pdf'
pp = PdfPages(plot_filename)
pp.savefig()
pp.close()
print "Saved plot: ", plot_filename
plt.cla()
def cutToTime(self, time, vals, cutoff):
new_time = [t for t in time if t <= cutoff]
new_vals = [v for (v,t) in zip(vals, time) if t <= cutoff]
return new_time, new_vals
def recordData(self, xvals, yvals, filename):
try:
os.remove(filename)
except OSError:
pass
with open(filename, 'w') as f:
for x, y in zip(xvals, yvals):
f.write('{0}, {1}\n'.format(x,y))