def get_flumina_latencies_throughputs(flumina_dirs):
flumina_dirs = list(flumina_dirs)
flumina_latencies = [
results.get_erlang_latencies(dir) for dir in flumina_dirs
]
flumina_throughputs = [
results.get_erlang_throughput(dir) for dir in flumina_dirs
]
return flumina_latencies, flumina_throughputs
def get_flumina_latencies_throughputs(flumina_dirs,
experiment='value-barrier'):
flumina_dirs = list(flumina_dirs)
flumina_latencies = [
results.get_erlang_latencies(dir, experiment) for dir in flumina_dirs
]
flumina_throughputs = [
results.get_erlang_throughput(dir) for dir in flumina_dirs
]
return flumina_latencies, flumina_throughputs
def plot_scaleup(erlang_subdirs, flink_subdirs, output_file):
erlang_latencies = [
results.get_erlang_latencies(subdir) for subdir in erlang_subdirs
]
erlang_latencies_mean = [p50 for p10, p50, p90 in erlang_latencies]
erlang_latencies_diff_10 = [
p50 - p10 for p10, p50, p90 in erlang_latencies
]
erlang_latencies_diff_90 = [
p90 - p50 for p10, p50, p90 in erlang_latencies
]
erlang_throughputs = [
results.get_erlang_throughput(subdir) for subdir in erlang_subdirs
]
flink_latencies = [
results.get_flink_latencies(subdir) for subdir in flink_subdirs
]
flink_latencies_mean = [p50 for p10, p50, p90 in flink_latencies]
flink_latencies_diff_10 = [p50 - p10 for p10, p50, p90 in flink_latencies]
flink_latencies_diff_90 = [p90 - p50 for p10, p50, p90 in flink_latencies]
flink_throughputs = [
results.get_flink_throughput(subdir) for subdir in flink_subdirs
]
plt.rcParams.update({'font.size': 18})
fig, ax = plt.subplots()
ax.set_xlabel('Throughput (events/ms)')
ax.set_ylabel('Latency (ms)')
plt.yscale('log')
ax.errorbar(erlang_throughputs,
erlang_latencies_mean,
[erlang_latencies_diff_10, erlang_latencies_diff_90],
linestyle='-',
marker='o',
label='Flumina',
linewidth=1,
capthick=1,
capsize=3,
color='tab:blue')
ax.errorbar(flink_throughputs,
flink_latencies_mean,
[flink_latencies_diff_10, flink_latencies_diff_90],
linestyle='--',
marker='^',
label='Flink',
linewidth=1,
capthick=1,
capsize=3,
color='tab:red')
ax.legend()
plt.tight_layout()
#plt.savefig(output_file)
plt.show()
def main():
parser = argparse.ArgumentParser(description='Run Flink experiments')
group = parser.add_mutually_exclusive_group()
group.add_argument('-s', '--suite', help='Run the given experiment suite')
group.add_argument('-l',
'--list',
help='List experiment suites',
action='store_true')
group.add_argument(
'-f',
'--flink-results',
help='Process Flink results from the given output directory')
group.add_argument(
'-e',
'--erlang-results',
help='Process Erlang results from the given output directory')
args = parser.parse_args()
if args.list:
for name, suite in suites.items():
print('\n\t'.join([f'{name}:\n' + '=' * (1 + len(name))] +
[str(exp) for exp in suite.experiments]) + '\n')
exit(0)
if args.flink_results is not None:
p10, p50, p90 = results.get_flink_latencies(args.flink_results)
throughput = results.get_flink_throughput(args.flink_results)
network_data = results.get_network_data(
args.flink_results) / 1024.0 / 1024.0
print(f'Latency percentiles (ms): {p10:.0f} {p50:.0f} {p90:.0f}')
print(f'Mean throughput (events/ms): {throughput}')
print(f'Network data (MB): {network_data:0.1f}')
exit(0)
if args.erlang_results is not None:
p10, p50, p90 = results.get_erlang_latencies(args.erlang_results)
throughput = results.get_erlang_throughput(args.erlang_results)
network_data = results.get_network_data(
args.erlang_results) / 1024.0 / 1024.0
print(f'Latency percentiles (ms): {p10:.0f} {p50:.0f} {p90:.0f}')
print(f'Mean throughput (events/ms): {throughput}')
print(f'Network data (MB): {network_data:0.1f}')
exit(0)
if args.suite not in suites:
parser.print_usage()
exit(1)
suites[args.suite].run()
def plot_hb_ratio_scaleup(erlang_dir, flink_dir, output_file):
ratios = [1, 2, 5, 10, 100, 1000, 10_000]
erlang_subdirs = [
path.join(erlang_dir,
f'ab_exp5_15_10000_{hb_ratio}_5_optimizer_greedy')
for hb_ratio in ratios
]
flink_subdirs = [
path.join(flink_dir, f'n5_r15_q10000_h{hb_ratio}')
for hb_ratio in ratios
]
erlang_latencies = [
results.get_erlang_latencies(subdir) for subdir in erlang_subdirs
]
erlang_latencies_mean = [p50 for p10, p50, p90 in erlang_latencies]
erlang_latencies_diff_10 = [
p50 - p10 for p10, p50, p90 in erlang_latencies
]
erlang_latencies_diff_90 = [
p90 - p50 for p10, p50, p90 in erlang_latencies
]
erlang_network_data = [
results.get_network_data(subdir) / (1024.0 * 1024.0)
for subdir in erlang_subdirs
]
flink_latencies = [
results.get_flink_latencies(subdir) for subdir in flink_subdirs
]
flink_latencies_mean = [p50 for p10, p50, p90 in flink_latencies]
flink_latencies_diff_10 = [p50 - p10 for p10, p50, p90 in flink_latencies]
flink_latencies_diff_90 = [p90 - p50 for p10, p50, p90 in flink_latencies]
flink_network_data = [
results.get_network_data(subdir) / (1024.0 * 1024.0)
for subdir in flink_subdirs
]
plt.rcParams.update({'font.size': 18})
fig, ax = plt.subplots()
ax.set_xlabel('Network data (MB)')
ax.set_ylabel('Latency (ms)')
plt.xscale('log')
plt.yscale('log')
plt.xticks(erlang_network_data, [
'' if item < 0.7 else f'{item:.1f}' if item < 100.0 else f'{item:.0f}'
for item in erlang_network_data
],
rotation=-60)
ax.errorbar(erlang_network_data,
erlang_latencies_mean,
[erlang_latencies_diff_10, erlang_latencies_diff_90],
linestyle='-',
marker='o',
label='Flumina',
linewidth=1,
capthick=1,
capsize=3,
color='tab:blue')
ax.errorbar(flink_network_data,
flink_latencies_mean,
[flink_latencies_diff_10, flink_latencies_diff_90],
linestyle='--',
marker='^',
label='Flink',
linewidth=1,
capthick=1,
capsize=3,
color='tab:red')
ax.legend()
plt.tight_layout()
plt.savefig(output_file)
def main():
parser = argparse.ArgumentParser(description='Run Flink experiments')
group = parser.add_mutually_exclusive_group()
group.add_argument('-s', '--suite', help='Run the given experiment suite')
group.add_argument('-e', '--experiment', help='Run the given experiment')
group.add_argument('-l', '--list', help='List experiment suites', action='store_true')
group.add_argument('-k', '--flink-results', help='Process Flink results from the given output directory')
group.add_argument('-f', '--flumina-results', help='Process Flumina results from the given output directory')
parser.add_argument('--flink-workers', help='File containing a list of Flink worker hostnames')
parser.add_argument('--total-values', type=int)
parser.add_argument('--value-nodes', type=int)
parser.add_argument('--value-rate', type=float)
parser.add_argument('--vb-ratio', type=int)
parser.add_argument('--hb-ratio', type=int)
parser.add_argument('--total-pageviews', type=int)
parser.add_argument('--total-users', type=int, default=2)
parser.add_argument('--pageview-parallelism', type=int)
parser.add_argument('--pageview-rate', type=float)
parser.add_argument('--sequential', action='store_true')
parser.add_argument('--manual', help='Run Flink with a manually implemented Flumina-like synchronization',
action='store_true')
parser.add_argument('--attempts', type=int, default=1)
parser.add_argument('--rmi-host', help='Host that is running the Java RMI registry')
args = parser.parse_args()
if args.list:
for name, suite in suites.items():
print('\n\t'.join(
[f'{name}:\n' + '=' * (1 + len(name))]
+ [str(exp) for exp in suite.experiments]) + '\n')
exit(0)
if args.flink_results is not None:
p10, p50, p90 = results.get_flink_latencies(args.flink_results)
throughput = results.get_flink_throughput(args.flink_results)
#network_data = results.get_network_data(args.flink_results) / 1024.0 / 1024.0
print(f'Latency percentiles (ms): {p10:.0f} {p50:.0f} {p90:.0f}')
print(f'Mean throughput (events/ms): {throughput}')
#print(f'Network data (MB): {network_data:0.1f}')
exit(0)
if args.flumina_results is not None:
p10, p50, p90 = results.get_erlang_latencies(args.erlang_results)
throughput = results.get_erlang_throughput(args.erlang_results)
network_data = results.get_network_data(args.erlang_results) / 1024.0 / 1024.0
print(f'Latency percentiles (ms): {p10:.0f} {p50:.0f} {p90:.0f}')
print(f'Mean throughput (events/ms): {throughput}')
print(f'Network data (MB): {network_data:0.1f}')
exit(0)
if args.experiment is not None:
if args.experiment.startswith("value-barrier"):
exps = [ValueBarrierEC2(args.value_nodes, args.total_values, args.value_rate, args.vb_ratio, args.hb_ratio,
manual=args.manual, sequential=args.sequential, attempt=a)
for a in range(1, args.attempts + 1)]
elif args.experiment.startswith("pageview"):
exps = [PageViewEC2(args.total_pageviews, args.total_users, args.pageview_parallelism, args.pageview_rate,
manual=args.manual, sequential=args.sequential, attempt=a)
for a in range(1, args.attempts + 1)]
elif args.experiment.startswith("fraud-detection"):
exps = [
FraudDetectionEC2(args.value_nodes, args.total_values, args.value_rate, args.vb_ratio, args.hb_ratio,
manual=args.manual, attempt=a)
for a in range(1, args.attempts + 1)]
ExperimentSuite(args.experiment, exps).run(args)
exit(0)
if args.suite not in suites:
parser.print_usage()
exit(1)
suites[args.suite].run(args)