def plot_train_qlen(self, input_dir, plt_name, traffic_files, algorithm):
algo = algorithm[0]
conf = algorithm[1]
# folders = glob.glob('%s/%s_*' % (input_dir, conf['pre']))
bb = {}
for tf in traffic_files:
for e in range(self.epochs):
bb['%s_%s' % (algo, e)] = []
input_file = input_dir + \
'/%s_%d/%s/qlen.txt' % (conf['pre'], e, tf)
results = self.get_qlen_stats(input_file, conf['sw'])
avg_qlen = float(results['avg_qlen'])
print("%s %s Epoch %d: Avg Qlen %f" % (algo, tf, e, avg_qlen))
bb['%s_%s' % (algo, e)].append(avg_qlen)
p_bar = []
p_legend = []
for i in range(self.epochs):
p_bar.append(bb['%s_%d' % (algo, i)][0])
p_legend.append('Epoch %i' % i)
print("%s: Total Average Qlen: %f" % (tf, avg(p_bar)))
plt.plot(p_bar)
x_val = list(range(self.epochs + 1))
if self.epochs > 100:
x_step = x_val[0::(self.epochs / 10)]
plt.xticks(x_step)
plt.xlabel('Epoch')
plt.ylabel('Average Queue Length')
axes = plt.gca()
axes.set_ylim([0, self.max_queue])
plt.savefig("%s/%s_%s" % (self.plt_dir, plt_name, tf))
plt.gcf().clear()
def plot_train_bw(self, input_dir, plt_name, traffic_files, algorithm):
algo = algorithm[0]
conf = algorithm[1]
fbb = self.num_ifaces * self.max_bw
# folders = glob.glob('%s/%s_*' % (input_dir, conf['pre']))
bb = {}
for tf in traffic_files:
for e in range(self.epochs):
bb['%s_%s' % (algo, e)] = []
input_file = input_dir + \
'/%s_%d/%s/rate_summary.json' % (conf['pre'], e, tf)
results = self.get_bw_dict(input_file)
avg_bw = float(results['avg_bw'])
print("%s %s Epoch %d: Avg BW %f" % (algo, tf, e, avg_bw))
bb['%s_%s' % (algo, e)].append(avg_bw / fbb / 2)
p_bar = []
p_legend = []
for i in range(self.epochs):
p_bar.append(bb['%s_%d' % (algo, i)][0])
p_legend.append('Epoch %i' % i)
print("%s: Total Average Bandwidth: %f" % (tf, avg(p_bar)))
plt.plot(p_bar)
x_val = list(range(self.epochs + 1))
if self.epochs > 100:
x_step = x_val[0::(self.epochs / 10)]
plt.xticks(x_step)
plt.xlabel('Epoch')
plt.ylabel('Normalized Average Bisection Bandwidth')
axes = plt.gca()
axes.set_ylim([-0.1, 1.1])
plt.savefig("%s/%s_%s" % (self.plt_dir, plt_name, tf))
plt.gcf().clear()
def evolving_average(self, input_list, n=100):
avgreward = []
summation = avg(input_list[:n - 1])
i = n
for r in input_list[n:]:
summation += float(r)
avgreward.append(summation / float(i))
i += 1
return avgreward
def get_bw_stats(self, input_file, pat_iface):
pat_iface = re.compile(pat_iface)
data = read_list(input_file)
rate = {}
column = 3
for row in data:
try:
ifname = row[1]
except Exception as e:
break
if pat_iface.match(ifname):
if ifname not in rate:
rate[ifname] = []
try:
rate[ifname].append(float(row[column]) * 8.0 / (1 << 20))
except Exception as e:
break
vals = {}
avg_bw = []
match = 0
avg_bw_iface = {}
filtered_bw = {}
for iface, bws in rate.items():
rate[iface] = self.moving_average(bws[10:-10], 100)
avg_bw.append(avg(bws[10:-10]))
avg_bw_iface[iface] = avg(bws[10:-10])
match += 1
# Update the num of interfaces to reflect true matches
# TODO: This is a hack. Remove.
self.num_ifaces = match
total_bw = list(itertools.chain.from_iterable(rate.values()))
vals["avg_bw_iface"] = avg_bw_iface
vals["avg_bw"] = fsum(avg_bw)
vals["median_bw"] = np.median(total_bw)
vals["max_bw"] = max(total_bw)
vals["stdev_bw"] = stdev(total_bw)
return vals, rate
def get_qlen_stats(self, input_file, pat_iface):
data = read_list(input_file)
pat_iface = re.compile(pat_iface)
qlen = {}
for row in data:
try:
ifname = row[0]
except Exception as e:
break
if pat_iface.match(ifname):
if ifname not in qlen:
qlen[ifname] = []
try:
qlen[ifname].append(int(row[2]))
except Exception as e:
break
vals = {}
qlens = []
avg_qlen_iface = {}
full_qlen_iface = {}
for iface, queues in qlen.items():
qlens.append(queues)
avg_qlen_iface[iface] = avg(queues)
full_qlen_iface[iface] = queues
# qlens = map(float, col(2, data))[10:-10]
qlens = list(itertools.chain.from_iterable(qlens))
vals["full_qlen_iface"] = full_qlen_iface
vals["avg_qlen_iface"] = avg_qlen_iface
vals["avg_qlen"] = avg(qlens)
vals["median_qlen"] = np.median(qlens)
vals["max_qlen"] = max(qlens)
vals["stdev_qlen"] = stdev(qlens)
vals["xcdf_qlen"], vals["ycdf_qlen"] = cdf(qlens)
return vals
def plot_train_qlen_iface(self, input_dir, plt_name, traffic_files,
algorithm):
self.check_plt_dir(plt_name)
algo = algorithm[0]
conf = algorithm[1]
# folders = glob.glob('%s/%s_*' % (input_dir, conf['pre']))
bb = {}
for tf in traffic_files:
for epoch in range(self.epochs):
bb['%s_%s' % (algo, epoch)] = {}
input_file = input_dir + \
'/%s_%d/%s/qlen.txt' % (conf['pre'], epoch, tf)
results = self.get_qlen_stats(input_file, conf['sw'])
avg_qlen = results["avg_qlen_iface"]
print("Qlen Average: %s %s Epoch %d" % (algo, tf, epoch))
print(avg_qlen)
for iface, qlen in avg_qlen.items():
bb['%s_%s' % (algo, epoch)].setdefault(iface, []).append(
float(qlen))
for iface, bw in bb['%s_%s' % (algo, 0)].items():
p_bar = []
p_legend = []
for i in range(self.epochs):
p_bar.append(bb['%s_%d' % (algo, i)][iface][0])
p_legend.append('Epoch %i' % i)
print("Interface %s: Total Average Qlen: %f" %
(iface, avg(p_bar)))
plt.plot(p_bar, label=iface)
x_val = list(range(self.epochs + 1))
if self.epochs > 100:
x_step = x_val[0::(self.epochs / 10)]
plt.xticks(x_step)
plt.xlabel('Epoch')
plt.ylabel('Average Queue Length')
axes = plt.gca()
# axes.set_ylim([0, self.max_queue])
plt.legend(loc='center right')
plt.savefig("%s_%s" % (plt_name, tf))
plt.gcf().clear()
def plot_train_bw_iface(self, input_dir, plt_name, traffic_files,
algorithm):
self.check_plt_dir(plt_name)
algo = algorithm[0]
conf = algorithm[1]
fbb = self.max_bw
# folders = glob.glob('%s/%s_*' % (input_dir, conf['pre']))
bb = {}
for tf in traffic_files:
for epoch in range(self.epochs):
bb['%s_%s' % (algo, epoch)] = {}
input_file = input_dir + \
'/%s_%d/%s/rate_summary.json' % (conf['pre'], epoch, tf)
results = self.get_bw_dict(input_file)
avg_bw = results['avg_bw_iface']
print("Average BW %s %s Epoch %d" % (algo, tf, epoch))
print(avg_bw)
for iface, bw in avg_bw.iteritems():
bb['%s_%s' % (algo, epoch)].setdefault(iface, []).append(
float(bw) / fbb)
for iface, bw in bb['%s_%s' % (algo, 0)].iteritems():
p_bar = []
p_legend = []
for i in range(self.epochs):
p_bar.append(bb['%s_%d' % (algo, i)][iface][0])
p_legend.append('Epoch %i' % i)
print("Interface %s: Total Average Bandwidth: %f" %
(iface, avg(p_bar)))
plt.plot(p_bar, label=iface)
x_val = list(range(self.epochs + 1))
if self.epochs > 100:
x_step = x_val[0::(self.epochs / 10)]
plt.xticks(x_step)
plt.xlabel('Epoch')
plt.ylabel('Normalized Average Bisection Bandwidth')
axes = plt.gca()
axes.set_ylim([-0.1, 1.1])
plt.legend(loc='center right')
plt.savefig("%s_%s" % (plt_name, tf))
plt.gcf().clear()
def plot_effective_bw(self, input_dir, plt_name, traffic_files, algorithm):
algo = algorithm[0]
conf = algorithm[1]
# folders = glob.glob('%s/%s_*' % (input_dir, conf['pre']))
bb = {}
hosts = ["h1", "h2", "h3", "h4"]
for tf in traffic_files:
for epoch in range(self.epochs):
bb['%s_%s' % (algo, epoch)] = {}
for host in hosts:
input_file = input_dir + \
'/%s_%d/%s/%s.out' % (conf['pre'], epoch, tf, host)
eff_bw_stat = self.get_eff_bw_stats(input_file)
avg_eff_bw = np.median(eff_bw_stat)
print("Effective Bandwidth Average: %s %s Epoch %d" %
(algo, tf, epoch))
print(avg_eff_bw)
bb['%s_%s' % (algo, epoch)][host] = avg_eff_bw
for host, avg_eff_bw in bb['%s_%s' % (algo, 0)].items():
p_bar = []
p_legend = []
for i in range(self.epochs):
p_bar.append(bb['%s_%d' % (algo, i)][host])
p_legend.append('Epoch %i' % i)
print("Host %s: Total Average Effective Bandwidth: %f" %
(host, avg(p_bar)))
plt.plot(p_bar, label=host)
x_val = list(range(self.epochs + 1))
if self.epochs > 100:
x_step = x_val[0::(self.epochs / 10)]
plt.xticks(x_step)
plt.xlabel('Epoch')
plt.ylabel('Average Effective Bandwidth (bytes/sec)')
axes = plt.gca()
# axes.set_ylim([0, self.max_queue])
plt.legend(loc='center right')
plt.savefig("%s/%s_%s" % (self.plt_dir, plt_name, tf))
plt.gcf().clear()
