def sec5_evaluation_microbenchmarks():
# main example, inbox250
basic = plot.PlotBasic('Scenario1', 'Scenario1', 180, verbose_logs=True, run_simulation=args.run_simulation, plot_all=False)
basic.thesismain_micro_latency()
basic.thesismain_micro_inbox_solidification_buffer()
basic.thesismain_micro_throughput(annotation='Scenario1')
healthor = plot.PlotV1('Scenario1V1', 'Scenario1V1', 180, verbose_logs=True, run_simulation=args.run_simulation, plot_all=False)
healthor.thesismain_micro_inbox_solidification_buffer()
healthor.thesismain_micro_latency()
healthor.thesismain_micro_available_processing()
healthor.thesismain_micro_throughput()
# Scenario2
# main example, inbox250
basic = plot.PlotBasic('Scenario2', 'Scenario1', 180, verbose_logs=True, run_simulation=args.run_simulation, plot_all=False)
print_latency_percentile_node(basic, 'node[2]')
print_latency_percentile_node(basic, 'node[6]')
plot_latency_percentile(basic, '', plot=False) # only print network latency percentile
basic.thesismain_micro_latency()
basic.thesismain_micro_inbox_solidification_buffer('Scenario2')
basic.thesismain_micro_throughput('Scenario2')
healthor = plot.PlotV1('Scenario2V1', 'Scenario1V1', 180, verbose_logs=True, run_simulation=args.run_simulation, plot_all=False)
print_latency_percentile_node(healthor, 'node[2]')
print_latency_percentile_node(healthor, 'node[6]')
plot_latency_percentile(healthor, '', plot=False) # only print network latency percentile
healthor.thesismain_micro_inbox_solidification_buffer()
healthor.thesismain_micro_latency()
healthor.thesismain_micro_available_processing(annotation='Scenario2')
healthor.thesismain_micro_throughput()
# inbox test series
plot.PlotBasic('Scenario2Inbox100', 'Scenario1', 180, verbose_logs=True, run_simulation=args.run_simulation, plot_all=True)
plot.PlotV1('Scenario2V1Inbox100', 'Scenario1V1', 180, verbose_logs=True, run_simulation=args.run_simulation, plot_all=True)
plot.PlotBasic('Scenario2Inbox500', 'Scenario1', 180, verbose_logs=True, run_simulation=args.run_simulation, plot_all=True)
plot.PlotV1('Scenario2V1Inbox500', 'Scenario1V1', 180, verbose_logs=True, run_simulation=args.run_simulation, plot_all=True)
plot.PlotBasic('Scenario2Inbox1000', 'Scenario1', 180, verbose_logs=True, run_simulation=args.run_simulation, plot_all=True)
plot.PlotV1('Scenario2V1Inbox1000', 'Scenario1V1', 180, verbose_logs=True, run_simulation=args.run_simulation, plot_all=True)
plot.PlotBasic('Scenario2Inbox2000', 'Scenario1', 180, verbose_logs=True, run_simulation=args.run_simulation, plot_all=True)
plot.PlotV1('Scenario2V1Inbox2000', 'Scenario1V1', 180, verbose_logs=True, run_simulation=args.run_simulation, plot_all=True)
# outbox test series
healthor = plot.PlotV1('Scenario2V1Outbox100', 'Scenario1V1', 180, verbose_logs=True, run_simulation=args.run_simulation, plot_all=True)
healthor.plot_buffers(to_node='node[4]')
healthor.plot_buffers(to_node='node[6]')
healthor = plot.PlotV1('Scenario2V1', 'Scenario1V1', 180, verbose_logs=True, run_simulation=args.run_simulation, plot_all=True)
healthor.plot_buffers(to_node='node[4]')
healthor.plot_buffers(to_node='node[6]')
healthor = plot.PlotV1('Scenario2V1Outbox500', 'Scenario1V1', 180, verbose_logs=True, run_simulation=args.run_simulation, plot_all=True)
healthor.plot_buffers(to_node='node[4]')
healthor.plot_buffers(to_node='node[6]')
healthor = plot.PlotV1('Scenario2V1Outbox1000', 'Scenario1V1', 180, verbose_logs=True, run_simulation=args.run_simulation, plot_all=True)
healthor.plot_buffers(to_node='node[4]')
healthor.plot_buffers(to_node='node[6]')
healthor = plot.PlotV1('Scenario2V1Outbox2000', 'Scenario1V1', 180, verbose_logs=True, run_simulation=args.run_simulation, plot_all=True)
healthor.plot_buffers(to_node='node[4]')
healthor.plot_buffers(to_node='node[6]')
def centralization_score_sensitivity():
runs = [
'MacroBasicSensitivity100T102',
'MacroBasicSensitivity100T104',
'MacroBasicSensitivity100T98',
'MacroBasicSensitivity100T96',
'MacroBasicSensitivity100T94',
'MacroBasicSensitivity100T92',
'MacroHealthorSensitivity100T96',
'MacroHealthorSensitivity100T98',
'MacroHealthorSensitivity100T102',
'MacroHealthorSensitivity100T104',
'MacroHealthorSensitivity100T103',
'MacroHealthorSensitivity100T106',
'MacroHealthorSensitivity100T108',
'MacroHealthorSensitivity100T110',
'MacroHealthorSensitivity100T112',
'MacroHealthorSensitivity100T114',
'MacroHealthorSensitivity100T116',
'MacroHealthorSensitivity100T118',
'MacroHealthorSensitivity100T120',
'MacroHealthorSensitivity100T122',
'MacroHealthorSensitivity100T124',
'MacroHealthorSensitivity100T126',
'MacroHealthorSensitivity100T140',
'MacroHealthorSensitivity100T160',
'MacroHealthorSensitivity100T180',
'MacroHealthorSensitivity100T200',
]
for name in runs:
print('Plotting %s...' % name)
if name.startswith('MacroHealthor'):
network = 'DynamicNetworkV1'
p = plot.PlotV1(name, network, 180, verbose_logs=False, run_simulation=args.run_simulation, plot_all=True)
else:
network = 'DynamicNetworkBasic'
p = plot.PlotBasic(name, network, 180, verbose_logs=False, run_simulation=args.run_simulation, plot_all=True)
# only print the necessary values for the table
plot_throughput_mean(p, '', plot=False)
plot_latency_percentile(p, '', plot=False)
plot_egalitarian_gradient(p, '', plot=False)
del p
print('Plotting %s... done' % name)
print('---------------------------------------')
def utilization():
# utilization N=100
p = plot.PlotV1('MacroHealthorUtilizationN100T106', 'DynamicNetworkV1', 180, verbose_logs=False,
run_simulation=args.run_simulation, plot_all=False)
# only print the necessary values for the table
plot_throughput_mean(p, '', plot=False)
plot_latency_percentile(p, '', plot=False)
plot_egalitarian_gradient(p, '', plot=False)
runs = [
'MacroHealthorUtilizationN500T128',
'MacroHealthorUtilizationN500T126',
'MacroHealthorUtilizationN500T124',
'MacroHealthorUtilizationN1000T116',
'MacroHealthorUtilizationN1000T114',
'MacroHealthorUtilizationN1000T112',
'MacroHealthorUtilizationN2000T110',
'MacroHealthorUtilizationN2000T108',
'MacroHealthorUtilizationN2000T106',
]
for name in runs:
print('Plotting %s...' % name)
if name.startswith('MacroHealthor'):
network = 'DynamicNetworkV1'
p = plot.PlotV1(name, network, 180, verbose_logs=False, run_simulation=args.run_simulation, plot_all=True)
else:
network = 'DynamicNetworkBasic'
p = plot.PlotBasic(name, network, 180, verbose_logs=False, run_simulation=args.run_simulation, plot_all=True)
# only print the necessary values for the table
plot_throughput_mean(p, '', plot=False)
plot_latency_percentile(p, '', plot=False)
plot_egalitarian_gradient(p, '', plot=False)
del p
print('Plotting %s... done' % name)
print('---------------------------------------')
def test_series():
plot.PlotBasic('BasicInbox100', 'DynamicNetworkBasic', 180, verbose_logs=False, run_simulation=True)
plot.PlotBasic('BasicInbox100FFA', 'DynamicNetworkBasic', 180, verbose_logs=False, run_simulation=True)
plot.PlotV1('V1Inbox100Outbox50', 'DynamicNetworkV1', 180, verbose_logs=False, run_simulation=True)
plot.PlotV1('V1Inbox100Outbox100', 'DynamicNetworkV1', 180, verbose_logs=False, run_simulation=True)
plot.PlotV1('V1Inbox100Outbox200', 'DynamicNetworkV1', 180, verbose_logs=False, run_simulation=True)
plot.PlotBasic('BasicInbox500', 'DynamicNetworkBasic', 180, verbose_logs=False, run_simulation=True)
plot.PlotBasic('BasicInbox500FFA', 'DynamicNetworkBasic', 180, verbose_logs=False, run_simulation=True)
plot.PlotV1('V1Inbox500Outbox100', 'DynamicNetworkV1', 180, verbose_logs=False, run_simulation=True)
plot.PlotV1('V1Inbox500Outbox500', 'DynamicNetworkV1', 180, verbose_logs=False, run_simulation=True)
plot.PlotV1('V1Inbox500Outbox1000', 'DynamicNetworkV1', 180, verbose_logs=False, run_simulation=True)
plot.PlotBasic('BasicInbox100Generation90', 'DynamicNetworkBasic', 180, verbose_logs=False, run_simulation=True)
plot.PlotBasic('BasicInbox100Generation90FFA', 'DynamicNetworkBasic', 180, verbose_logs=False, run_simulation=True)
plot.PlotV1('V1Inbox100Outbox50Generation90', 'DynamicNetworkV1', 180, verbose_logs=False, run_simulation=True)
plot.PlotV1('V1Inbox100Outbox100Generation90', 'DynamicNetworkV1', 180, verbose_logs=False, run_simulation=True)
plot.PlotV1('V1Inbox100Outbox200Generation90', 'DynamicNetworkV1', 180, verbose_logs=False, run_simulation=True)
plot.PlotBasic('BasicInbox500Generation90', 'DynamicNetworkBasic', 180, verbose_logs=False, run_simulation=True)
plot.PlotBasic('BasicInbox500Generation90FFA', 'DynamicNetworkBasic', 180, verbose_logs=False, run_simulation=True)
plot.PlotV1('V1Inbox500Outbox100Generation90', 'DynamicNetworkV1', 180, verbose_logs=False, run_simulation=True)
plot.PlotV1('V1Inbox500Outbox500Generation90', 'DynamicNetworkV1', 180, verbose_logs=False, run_simulation=True)
plot.PlotV1('V1Inbox500Outbox1000Generation90', 'DynamicNetworkV1', 180, verbose_logs=False, run_simulation=True)
def sec5_evaluation_macrobenchmarks():
plot.PlotBasic('MacroBasic100', 'DynamicNetworkBasic', 180, verbose_logs=False, run_simulation=args.run_simulation, plot_all=False)
plot.PlotBasic('MacroFFA100', 'DynamicNetworkBasic', 180, verbose_logs=False, run_simulation=args.run_simulation, plot_all=False)
plot.PlotV1('MacroHealthor100', 'DynamicNetworkV1', 180, verbose_logs=False, run_simulation=args.run_simulation, plot_all=False)
plot.PlotBasic('MacroBasic500', 'DynamicNetworkBasic', 180, verbose_logs=False, run_simulation=args.run_simulation, plot_all=False)
plot.PlotBasic('MacroFFA500', 'DynamicNetworkBasic', 180, verbose_logs=False, run_simulation=args.run_simulation, plot_all=False)
plot.PlotV1('MacroHealthor500', 'DynamicNetworkV1', 180, verbose_logs=False, run_simulation=args.run_simulation, plot_all=False)
plot.PlotBasic('MacroBasic1000', 'DynamicNetworkBasic', 180, verbose_logs=False, run_simulation=args.run_simulation, plot_all=False)
plot.PlotBasic('MacroFFA1000', 'DynamicNetworkBasic', 180, verbose_logs=False, run_simulation=args.run_simulation, plot_all=False)
plot.PlotV1('MacroHealthor1000', 'DynamicNetworkV1', 180, verbose_logs=False, run_simulation=args.run_simulation, plot_all=False)
basic = plot.PlotBasic('MacroBasic2000', 'DynamicNetworkBasic', 180, verbose_logs=False, run_simulation=args.run_simulation, plot_all=False)
ffa = plot.PlotBasic('MacroFFA2000', 'DynamicNetworkBasic', 180, verbose_logs=False, run_simulation=args.run_simulation, plot_all=False)
healthor = plot.PlotV1('MacroHealthor2000', 'DynamicNetworkV1', 180, verbose_logs=False, run_simulation=args.run_simulation, plot_all=False)
# generate plots for macrobenchmark section
sec5_throughput(basic, ffa, healthor, legend=False)
sec5_egalitarian_gradient(basic, ffa, healthor, legend=True)
sec5_latency_percentiles(basic, ffa, healthor, legend=False)
plot.PlotBasic('MacroBasic5000', 'DynamicNetworkBasic', 180, verbose_logs=False, run_simulation=args.run_simulation, plot_all=False)
plot.PlotBasic('MacroFFA5000', 'DynamicNetworkBasic', 180, verbose_logs=False, run_simulation=args.run_simulation, plot_all=False)
plot.PlotV1('MacroHealthor5000', 'DynamicNetworkV1', 180, verbose_logs=False, run_simulation=args.run_simulation, plot_all=False)
return
# this can be utilized once all the runs are done (e.g. to populate the results table)
runs = [
# 'MacroBasic100',
# 'MacroFFA100',
# 'MacroHealthor100',
#
# 'MacroBasic500',
# 'MacroFFA500',
# 'MacroHealthor500',
#
# 'MacroBasic1000',
# 'MacroFFA1000',
# 'MacroHealthor1000',
#
# 'MacroBasic2000',
# 'MacroFFA2000',
# 'MacroHealthor2000',
# 'MacroBasic5000',
# 'MacroFFA5000',
# 'MacroHealthor5000',
]
# for name in runs:
# print('Plotting %s...' % name)
# if name.startswith('MacroHealthor'):
# network = 'DynamicNetworkV1'
# p = plot.PlotV1(name, network, 180, verbose_logs=False, run_simulation=False, plot_all=False)
# else:
# network = 'DynamicNetworkBasic'
# p = plot.PlotBasic(name, network, 180, verbose_logs=False, run_simulation=False, plot_all=False)
#
# # only print the necessary values for the table
# # plot_throughput_mean(p, '', plot=False)
# plot_latency_percentile(p, '', plot=False)
# # plot_egalitarian_gradient(p, '', plot=False)
# del p
# print('Plotting %s... done' % name)
# print('---------------------------------------')
# calculate values for 5000 FFA (too big to run normally)
def throughput_5000_special(path):
df = pd.read_csv(path)
all_nodes = list(filter(lambda x: x.startswith('node'), list(df)))
print(len(all_nodes))
# determine how many nodes were in sync and average
total_count = None
total_in_sync_nodes = None
for node in all_nodes:
count, division = np.histogram(df[node], bins=range(180))
# sum up for throughput average
if total_count is not None:
total_count += count
total_in_sync_nodes += (count > 0).astype(int)
else:
total_count = count
total_in_sync_nodes = (count > 0).astype(int)
# determine average
# print(total_count[:-11])
# print(total_in_sync_nodes[:-11])
total_count = total_count[:-11] / total_in_sync_nodes[:-11]
print('Throughput avg mean: %.2f' % np.mean(total_count))
# throughput_5000_special('msgs-FFA-5000.csv')
def latency_5000_special(path, percentile):
df = pd.read_csv(path)
all_nodes = list(filter(lambda x: x.startswith('node'), list(df)))
print(len(all_nodes))
difference_suffix = '_difference'
msgs_count = df.shape[0]
# determine which nodes CDF needs to be calculated
nodes_in_sync = []
# determine whether node was in sync (for most of the simulation)
for node in all_nodes:
msgs_node_missing = df[node].isnull().sum()
if msgs_node_missing < msgs_count * 0.05:
nodes_in_sync.append(node)
print('nodes in sync count', len(nodes_in_sync))
print('messages count', msgs_count)
# calculate latency for every message on every in sync node
# runtime: ~40s
for node in nodes_in_sync:
df[node + difference_suffix] = df[node].subtract(df['time'])
print(node, 'done')
#
nodes_difference_columns = [x + difference_suffix for x in nodes_in_sync]
percentile_name = 'percentile_' + str(percentile)
percentile_name_difference = percentile_name + '_difference'
percentile_df = pd.DataFrame(columns=[percentile_name_difference])
percentile_df[percentile_name_difference] = df[nodes_difference_columns].quantile(percentile, axis=1)
def calculate_cdf(d, n):
node_difference = n + '_difference'
stats_df = d.groupby([node_difference])[node_difference].agg('count').pipe(pd.DataFrame).rename(
columns={node_difference: 'frequency'})
stats_df['pdf'] = stats_df['frequency'] / sum(stats_df['frequency'])
stats_df['cdf'] = stats_df['pdf'].cumsum()
stats_df = stats_df.reset_index()
return stats_df
cdf_df = calculate_cdf(percentile_df, percentile_name)
# print 0.95 percentile latency of CDF
np_arr = cdf_df['cdf'].to_numpy()
m_index = np.where(np_arr >= 0.95)[0][0]
print('Percentile CDF %.2f 0.95th: %.2f' % (percentile, cdf_df[percentile_name_difference][m_index]))
def sec2():
basic = plot.PlotBasic('DynamicBasic100', 'DynamicNetworkBasic', 180, verbose_logs=False, run_simulation=True, plot_all=False)
ffa = plot.PlotBasic('DynamicBasic100FFA', 'DynamicNetworkBasic', 180, verbose_logs=False, run_simulation=True, plot_all=False)
sec2_egalitarian_gradient(basic, ffa)
sec2_throughput(basic, ffa)