def indexMultiple(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '--masterServiceThreads 1'
if cluster_args['timeout'] < 2100:
cluster_args['timeout'] = 200
# Ensure atleast 15 hosts for optimal performance
if options.num_servers == None:
cluster_args['num_servers'] = len(hosts)
# use a maximum of 10 secondary keys
if len(hosts) <= 10:
# TODO(ankitak): hack until synchronization bug in write RPC handler
# in MasterService is resolved. This bug prevents us from using more
# than 1 MasterSerivice thread. However, we need to use more than 1
# service thread, otherwise if a tablet and its corresponding
# indexlet end up on the same server, we will have a deadlock.
# For now, make sure that we never wrap around the server list
# Once the bug is resolved, we should be able to use len(hosts)
# for numIndexes
client_args['--numIndexes'] = len(hosts) - 1
else:
client_args['--numIndexes'] = 10
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end='')
def workloadDist(name, options, cluster_args, client_args):
if not options.extract:
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '-t 2000'
cluster_args['disjunct'] = True
cluster.run(client='%s/apps/ClusterPerf %s %s' %
(config.hooks.get_remote_obj_path(),
flatten_args(client_args), name),
**cluster_args)
if options.fullSamples:
import gzip
with gzip.open('logs/latest/rcdf.data.gz', 'wb') as rcdf_file:
print_rcdf_from_log_samples(rcdf_file)
with gzip.open('logs/latest/samples.data.gz', 'wb') as samples_file:
print_samples_from_log(samples_file)
else:
print("# Cumulative distribution latencies for operations specified by\n"
"# the benchmark.\n#\n"
"# Generated by 'clusterperf.py %s'\n#\n"
"# Time (usec) Cum. Fraction\n"
"#---------------------------"
% (name))
if (options.rcdf):
print_rcdf_from_log()
else:
print_cdf_from_log()
def indexReadDist(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '--maxCores 2 --totalMasterMemory 1500'
if cluster_args['timeout'] < 200:
cluster_args['timeout'] = 200
if options.num_servers == None:
cluster_args['num_servers'] = len(getHosts())
if '--count' not in client_args:
client_args['--count'] = 10000
if '--numObjects' not in client_args:
client_args['--numObjects'] = 1000000
if '--warmup' not in client_args:
client_args['--warmup'] = 100
cluster.run(
client='%s/apps/ClusterPerf %s %s' %
(config.hooks.get_remote_obj_path(), flatten_args(client_args), name),
**cluster_args)
print("# Cumulative distribution of time for a single client to read\n"
"# %d %d-byte objects to a table with one index and %d\n"
"# initial objects. Each object has two 30-byte keys and a 100\n"
"# byte value. Each line indicates that a given fraction of all\n"
"# reads took at most a given time to complete.\n"
"#\n"
"# Generated by 'clusterperf.py readDist'\n#\n"
"# Time (usec) Cum. Fraction\n"
"#---------------------------" %
(client_args['--count'], options.size, client_args['--numObjects']))
print_cdf_from_log()
def readLoaded(name, options, cluster_args, client_args):
if 'num_clients' not in cluster_args:
cluster_args['num_clients'] = 20
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name),
**cluster_args)
print(get_client_log(), end='')
def workloadDist(name, options, cluster_args, client_args):
if not options.extract:
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '-t 2000'
cluster_args['disjunct'] = True
cluster.run(client='%s/apps/ClusterPerf %s %s' %
(config.hooks.get_remote_obj_path(),
flatten_args(client_args), name),
**cluster_args)
if options.fullSamples:
import gzip
with gzip.open('logs/latest/rcdf.data.gz', 'wb') as rcdf_file:
print_rcdf_from_log_samples(rcdf_file)
with gzip.open('logs/latest/samples.data.gz', 'wb') as samples_file:
print_samples_from_log(samples_file)
else:
print(
"# Cumulative distribution latencies for operations specified by\n"
"# the benchmark.\n#\n"
"# Generated by 'clusterperf.py %s'\n#\n"
"# Time (usec) Cum. Fraction\n"
"#---------------------------" % (name))
if (options.rcdf):
print_rcdf_from_log()
else:
print_cdf_from_log()
def indexMultiple(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '--masterServiceThreads 1'
if cluster_args['timeout'] < 360:
cluster_args['timeout'] = 360
# Ensure atleast 15 hosts for optimal performance
if options.num_servers == None:
cluster_args['num_servers'] = len(hosts)
# use a maximum of 10 secondary keys
if len(hosts) <= 10:
# Hack until synchronization bug in write RPC handler
# in MasterService is resolved. This bug prevents us from using more
# than 1 MasterSerivice thread. However, we need to use more than 1
# service thread, otherwise if a tablet and its corresponding
# indexlet end up on the same server, we will have a deadlock.
# For now, make sure that we never wrap around the server list
# Once the bug is resolved, we should be able to use len(hosts)
# for numIndexes
client_args['--numIndexes'] = len(hosts) - 1
else:
client_args['--numIndexes'] = 10
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end='')
def indexReadDist(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '--masterServiceThreads 1 --totalMasterMemory 1500'
if cluster_args['timeout'] < 200:
cluster_args['timeout'] = 200
if options.num_servers == None:
cluster_args['num_servers'] = len(hosts)
if '--count' not in client_args:
client_args['--count'] = 10000
if '--numObjects' not in client_args:
client_args['--numObjects'] = 1000000
if '--warmup' not in client_args:
client_args['--warmup'] = 100
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), **cluster_args)
print("# Cumulative distribution of time for a single client to read\n"
"# %d %d-byte objects to a table with one index and %d\n"
"# initial objects. Each object has two 30-byte keys and a 100\n"
"# byte value. Each line indicates that a given fraction of all\n"
"# reads took at most a given time to complete.\n"
"#\n"
"# Generated by 'clusterperf.py readDist'\n#\n"
"# Time (usec) Cum. Fraction\n"
"#---------------------------"
% (client_args['--count'], options.size, client_args['--numObjects'] ))
print_cdf_from_log()
def readLoaded(name, options, cluster_args, client_args):
if 'num_clients' not in cluster_args:
cluster_args['num_clients'] = 20
cluster.run(client='%s/apps/ClusterPerf %s %s' %
(config.hooks.get_remote_obj_path(),
flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end='')
def broadcast(name, options, cluster_args, client_args):
if 'num_clients' not in cluster_args:
cluster_args['num_clients'] = 10
cluster.run(
client='%s/apps/ClusterPerf %s %s' %
(config.hooks.get_remote_obj_path(), flatten_args(client_args), name),
**cluster_args)
print(get_client_log(), end='')
def basic(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '-t 4000'
if cluster_args['timeout'] < 250:
cluster_args['timeout'] = 250
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end='')
def multiOp(name, options, cluster_args, client_args):
cluster_args['timeout'] = 100
if options.num_servers == None:
cluster_args['num_servers'] = len(hosts)
client_args['--numTables'] = cluster_args['num_servers'];
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), master_args='-d', **cluster_args)
print(get_client_log(), end='')
def basic(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '-t 4000'
if cluster_args['timeout'] < 250:
cluster_args['timeout'] = 250
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end='')
def readRandom(name, options, cluster_args, client_args):
cluster_args["timeout"] = 60
if "num_clients" not in cluster_args:
cluster_args["num_clients"] = 50
if options.num_servers == None:
cluster_args["num_servers"] = 10
client_args["--numTables"] = cluster_args["num_servers"]
cluster.run(client="%s/ClusterPerf %s %s" % (obj_path, flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end="")
def migrateLoaded(name, options, cluster_args, client_args):
if not options.extract:
clients = options.num_clients
servers = options.num_servers # len(getHosts()) - clients - 1
if servers < 4:
raise Exception('Not enough servers: only %d left' % servers)
if clients < 16:
print('!!! WARNING !!! Use 16 clients to ensure enough load for ' +
'real experiments !!! WARNING !!!',
file=sys.stderr)
cluster_args['num_servers'] = servers
# Use two backups per server for more disk bandwidth.
defaultTo(cluster_args, 'backup_disks_per_server', 2)
# Need lots of mem for big workload and migration.
defaultTo(cluster_args, 'master_args', '-t 18000 --segmentFrames 8192')
# Sixteen clients to try to generate enough load to keep things at
# about 90% load.
cluster_args['num_clients'] = clients
# Can take awhile due to fillup and migration.
if cluster_args['timeout'] < 300:
cluster_args['timeout'] = 300
# We're really interested in jitter on servers; better keep the clients
# off the server machines.
cluster_args['disjunct'] = True
# Can't default --workload this due to command line default...
# 1 million * 100 B ~= 100 MB table
defaultTo(client_args, '--numObjects', 1 * 1000 * 1000)
# Set clients up to keep server at 90% load.
defaultTo(
client_args, '--targetOps',
calculatePerClientTarget(client_args['--workload'], clients,
options.loadPct))
# Turn on timestamps on latency samples.
defaultTo(client_args, '--fullSamples', '')
name = 'readDistWorkload'
cluster.run(client='%s/apps/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name),
**cluster_args)
import gzip
with gzip.open('logs/latest/rcdf.data.gz', 'wb') as rcdf_file:
print_rcdf_from_log_samples(rcdf_file)
with gzip.open('logs/latest/samples.data.gz', 'wb') as samples_file:
print_samples_from_log(samples_file)
def multiOp(name, options, cluster_args, client_args):
if cluster_args['timeout'] < 100:
cluster_args['timeout'] = 100
if options.num_servers == None:
cluster_args['num_servers'] = len(hosts)
client_args['--numTables'] = cluster_args['num_servers'];
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name),
**cluster_args)
print(get_client_log(), end='')
def readRandom(name, options, cluster_args, client_args):
cluster_args['timeout'] = 60
if 'num_clients' not in cluster_args:
cluster_args['num_clients'] = 50
if options.num_servers == None:
cluster_args['num_servers'] = 10
client_args['--numTables'] = cluster_args['num_servers'];
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end='')
def readThroughput(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '-t 2000'
if cluster_args['timeout'] < 250:
cluster_args['timeout'] = 250
if 'num_clients' not in cluster_args:
cluster_args['num_clients'] = len(hosts) - cluster_args['num_servers']
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end='')
def readAllToAll(name, options, cluster_args, client_args):
cluster_args["backups_per_server"] = 0
cluster_args["replicas"] = 0
if "num_clients" not in cluster_args:
cluster_args["num_clients"] = len(hosts)
if options.num_servers == None:
cluster_args["num_servers"] = len(hosts)
client_args["--numTables"] = cluster_args["num_servers"]
cluster.run(client="%s/ClusterPerf %s %s" % (obj_path, flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end="")
def indexBasic(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '--maxCores 2 --totalMasterMemory 1500'
if cluster_args['timeout'] < 200:
cluster_args['timeout'] = 200
# Ensure at least 5 hosts for optimal performance
if options.num_servers == None:
cluster_args['num_servers'] = len(getHosts())
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end='')
def txCollision(name, options, cluster_args, client_args):
if cluster_args['timeout'] < 100:
cluster_args['timeout'] = 100
if options.num_servers == None:
cluster_args['num_servers'] = len(getHosts())
#client_args['--numTables'] = cluster_args['num_servers'];
if 'num_clients' not in cluster_args:
cluster_args['num_clients'] = 5
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end='')
def readRandom(name, options, cluster_args, client_args):
cluster_args['timeout'] = 60
if 'num_clients' not in cluster_args:
cluster_args['num_clients'] = 50
if options.num_servers == None:
cluster_args['num_servers'] = 10
client_args['--numTables'] = cluster_args['num_servers']
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name),
**cluster_args)
print(get_client_log(), end='')
def readRandom(name, options, cluster_args, client_args):
cluster_args['backups_per_server'] = 0
cluster_args['replicas'] = 0
if 'num_clients' not in cluster_args:
cluster_args['num_clients'] = 16
if options.num_servers == None:
cluster_args['num_servers'] = 1
client_args['--numTables'] = cluster_args['num_servers'];
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end='')
def multiOp(name, options, cluster_args, client_args):
if cluster_args['timeout'] < 100:
cluster_args['timeout'] = 100
if options.num_servers == None:
cluster_args['num_servers'] = len(getHosts())
client_args['--numTables'] = cluster_args['num_servers'];
cluster.run(client='%s/apps/ClusterPerf %s %s' %
(config.hooks.get_remote_obj_path(),
flatten_args(client_args), name),
**cluster_args)
print(get_client_log(), end='')
def multiOp(name, options, cluster_args, client_args):
if cluster_args['timeout'] < 100:
cluster_args['timeout'] = 100
if options.num_servers == None:
cluster_args['num_servers'] = len(getHosts())
client_args['--numTables'] = cluster_args['num_servers']
cluster.run(
client='%s/apps/ClusterPerf %s %s' %
(config.hooks.get_remote_obj_path(), flatten_args(client_args), name),
**cluster_args)
print(get_client_log(), end='')
def indexBasic(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '--masterServiceThreads 1 --totalMasterMemory 1500'
if cluster_args['timeout'] < 200:
cluster_args['timeout'] = 200
# Ensure at least 5 hosts for optimal performance
if options.num_servers == None:
cluster_args['num_servers'] = len(hosts)
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end='')
def readRandom(name, options, cluster_args, client_args):
cluster_args['backups_per_server'] = 0
cluster_args['replicas'] = 0
if 'num_clients' not in cluster_args:
cluster_args['num_clients'] = 16
if options.num_servers == None:
cluster_args['num_servers'] = 1
client_args['--numTables'] = cluster_args['num_servers'];
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end='')
def txCollision(name, options, cluster_args, client_args):
if cluster_args['timeout'] < 100:
cluster_args['timeout'] = 100
if options.num_servers == None:
cluster_args['num_servers'] = len(hosts)
#client_args['--numTables'] = cluster_args['num_servers'];
if 'num_clients' not in cluster_args:
cluster_args['num_clients'] = 5
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end='')
def indexBasic(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '--maxCores 2 --totalMasterMemory 1500'
if cluster_args['timeout'] < 200:
cluster_args['timeout'] = 200
# Ensure at least 5 hosts for optimal performance
if options.num_servers == None:
cluster_args['num_servers'] = len(getHosts())
cluster.run(client='%s/apps/ClusterPerf %s %s' %
(config.hooks.get_remote_obj_path(),
flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end='')
def readAllToAll(name, options, cluster_args, client_args):
cluster_args['backup_disks_per_server'] = 0
cluster_args['replicas'] = 0
if 'num_clients' not in cluster_args:
cluster_args['num_clients'] = len(getHosts())
if options.num_servers == None:
cluster_args['num_servers'] = len(getHosts())
client_args['--numTables'] = cluster_args['num_servers']
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name),
**cluster_args)
print(get_client_log(), end='')
def workloadDist(name, options, cluster_args, client_args):
if not options.extract:
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '-t 2000'
cluster_args['disjunct'] = True
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name),
**cluster_args)
if options.rcdf:
print_rcdf_from_log_samples()
else:
print_samples_from_log()
def readAllToAll(name, options, cluster_args, client_args):
cluster_args['backup_disks_per_server'] = 0
cluster_args['replicas'] = 0
if 'num_clients' not in cluster_args:
cluster_args['num_clients'] = len(getHosts())
if options.num_servers == None:
cluster_args['num_servers'] = len(getHosts())
client_args['--numTables'] = cluster_args['num_servers'];
cluster.run(client='%s/apps/ClusterPerf %s %s' %
(config.hooks.get_remote_obj_path(),
flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end='')
def workloadDist(name, options, cluster_args, client_args):
if not options.extract:
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '-t 2000'
cluster_args['disjunct'] = True
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name),
**cluster_args)
if options.rcdf:
print_rcdf_from_log_samples()
else:
print_samples_from_log()
def netBandwidth(name, options, cluster_args, client_args):
if "num_clients" not in cluster_args:
cluster_args["num_clients"] = 2 * len(config.hosts)
if options.num_servers == None:
cluster_args["num_servers"] = cluster_args["num_clients"]
if cluster_args["num_servers"] > len(config.hosts):
cluster_args["num_servers"] = len(config.hosts)
if options.size != None:
client_args["--size"] = options.size
else:
client_args["--size"] = 1024 * 1024
cluster.run(client="%s/ClusterPerf %s %s" % (obj_path, flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end="")
def indexBasic(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '--masterServiceThreads 1'
if cluster_args['timeout'] < 200:
cluster_args['timeout'] = 200
# Ensure at least 5 hosts for optimal performance
if options.num_servers == None:
cluster_args['num_servers'] = len(hosts)
# using 20GB for servers so that we don't run out of memory when inserting
# 10 million objects/index entries
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end='')
def readDist(name, options, cluster_args, client_args):
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name),
**cluster_args)
print("# Cumulative distribution of time for a single client to read a\n"
"# single %d-byte object from a single server. Each line indicates\n"
"# that a given fraction of all reads took at most a given time\n"
"# to complete.\n"
"# Generated by 'clusterperf.py readDist'\n#\n"
"# Time (usec) Cum. Fraction\n"
"#---------------------------"
% options.size)
print_cdf_from_log()
def readDistRandom(name, options, cluster_args, client_args):
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name),
**cluster_args)
print("# Cumulative distribution of time for a single client to read a\n"
"# random %d-byte object from a single server. Each line indicates\n"
"# that a given fraction of all reads took at most a given time\n"
"# to complete.\n"
"# Generated by 'clusterperf.py readDist'\n#\n"
"# Time (usec) Cum. Fraction\n"
"#---------------------------"
% options.size)
print_cdf_from_log()
def netBandwidth(name, options, cluster_args, client_args):
if 'num_clients' not in cluster_args:
cluster_args['num_clients'] = 2*len(config.hosts)
if options.num_servers == None:
cluster_args['num_servers'] = cluster_args['num_clients']
if cluster_args['num_servers'] > len(config.hosts):
cluster_args['num_servers'] = len(config.hosts)
if options.size != None:
client_args['--size'] = options.size
else:
client_args['--size'] = 1024*1024;
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end='')
def echoWorkload(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '-t 4000'
if cluster_args['timeout'] < 250:
cluster_args['timeout'] = 250
cluster_args['replicas'] = 0
if options.num_servers == None:
cluster_args['num_servers'] = 1
cluster.run(
client='%s/apps/ClusterPerf %s %s' %
(config.hooks.get_remote_obj_path(), flatten_args(client_args), name),
**cluster_args)
print("# Generated by 'clusterperf.py echoWorkload'\n#\n")
print_percentiles_from_logs()
def migrateLoaded(name, options, cluster_args, client_args):
if not options.extract:
clients = 16
servers = len(getHosts()) - clients - 1
if servers < 4:
raise Exception('Not enough servers: only %d left' % servers)
cluster_args['num_servers'] = servers
# Use two backups per server for more disk bandwidth.
defaultTo(cluster_args, 'backup_disks_per_server', 2)
# Need lots of mem for big workload and migration.
defaultTo(cluster_args, 'master_args',
'-t 18000 --segmentFrames 8192')
# Sixteen clients to try to generate enough load to keep things at
# about 90% load.
cluster_args['num_clients'] = clients
# Can take awhile due to fillup and migration.
if cluster_args['timeout'] < 300:
cluster_args['timeout'] = 300
# We're really interested in jitter on servers; better keep the clients
# off the server machines.
cluster_args['disjunct'] = True
# Can't default --workload this due to command line default...
# 1 million * 100 B ~= 100 MB table
defaultTo(client_args, '--numObjects', 1 * 1000 * 1000)
# Set clients up to keep server at 90% load.
defaultTo(client_args, '--targetOps',
calculatePerClientTarget(
client_args['--workload'], clients,
options.loadPct))
name = 'readDistWorkload'
cluster.run(client='%s/apps/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name),
**cluster_args)
import gzip
with gzip.open('logs/latest/rcdf.data.gz', 'wb') as rcdf_file:
print_rcdf_from_log_samples(rcdf_file)
with gzip.open('logs/latest/samples.data.gz', 'wb') as samples_file:
print_samples_from_log(samples_file)
def netBandwidth(name, options, cluster_args, client_args):
if 'num_clients' not in cluster_args:
cluster_args['num_clients'] = 2*len(config.hosts)
if options.num_servers == None:
cluster_args['num_servers'] = cluster_args['num_clients']
if cluster_args['num_servers'] > len(config.hosts):
cluster_args['num_servers'] = len(config.hosts)
if options.size != None:
client_args['--size'] = options.size
else:
client_args['--size'] = 1024*1024;
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end='')
def readThroughput(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '-t 2000'
if cluster_args['timeout'] < 250:
cluster_args['timeout'] = 250
if 'num_clients' not in cluster_args:
cluster_args['num_clients'] = len(getHosts()) - cluster_args['num_servers']
if cluster_args['num_clients'] < 2:
print("Not enough machines in the cluster to run the '%s' benchmark"
% name)
print("Need at least 2 machines in this configuration")
return
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end='')
def indexScalability(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '--masterServiceThreads 2'
if cluster_args['timeout'] < 100:
cluster_args['timeout'] = 100
cluster_args['backups_per_server'] = 0
cluster_args['replicas'] = 0
# Ensure at least 15 hosts for optimal performance
if options.num_servers == None:
cluster_args['num_servers'] = len(hosts)
if 'num_clients' not in cluster_args:
cluster_args['num_clients'] = 10
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end='')
def readThroughput(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '-t 2000'
if cluster_args['timeout'] < 250:
cluster_args['timeout'] = 250
if 'num_clients' not in cluster_args:
cluster_args['num_clients'] = len(getHosts()) - cluster_args['num_servers']
if cluster_args['num_clients'] < 2:
print("Not enough machines in the cluster to run the '%s' benchmark"
% name)
print("Need at least 2 machines in this configuration")
return
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end='')
def writeDist(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '-t 2000'
cluster_args['disjunct'] = True
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name),
**cluster_args)
print("# Cumulative distribution of time for a single client to write a\n"
"# single %d-byte object from a single server. Each line indicates\n"
"# that a given fraction of all writes took at most a given time\n"
"# to complete.\n"
"# Generated by 'clusterperf.py %s'\n#\n"
"# Time (usec) Cum. Fraction\n"
"#---------------------------"
% (options.size, name))
print_cdf_from_log()
def workloadDist(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '-t 2000'
cluster_args['disjunct'] = True
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name),
**cluster_args)
print("# Cumulative distribution latencies for operations specified by\n"
"# the benchmark.\n#\n"
"# Generated by 'clusterperf.py %s'\n#\n"
"# Time (usec) Cum. Fraction\n"
"#---------------------------" % (name))
if (options.rcdf):
print_rcdf_from_log()
else:
print_cdf_from_log()
def txTpccLatency(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = ' -t 10000 '
if 'backup_args' not in cluster_args:
cluster_args[
'backup_args'] = ' --segmentFrames 10000 --maxNonVolatileBuffers 8'
if cluster_args['timeout'] < 150:
cluster_args['timeout'] = 150
#cluster_args['disjunct'] = True
if options.num_servers == None:
cluster_args['num_servers'] = 4
if 'num_clients' not in cluster_args:
cluster_args['num_clients'] = 1
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name),
**cluster_args)
print(get_client_log(), end='')
def workloadDist(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '-t 2000'
cluster_args['disjunct'] = True
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name),
**cluster_args)
print("# Cumulative distribution latencies for operations specified by\n"
"# the benchmark.\n#\n"
"# Generated by 'clusterperf.py %s'\n#\n"
"# Time (usec) Cum. Fraction\n"
"#---------------------------"
% (name))
if (options.rcdf):
print_rcdf_from_log()
else:
print_cdf_from_log()
def transactionThroughput(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '-t 2000'
if cluster_args['timeout'] < 250:
cluster_args['timeout'] = 250
if 'num_clients' not in cluster_args:
cluster_args['num_clients'] = len(getHosts()) - cluster_args['num_servers']
if cluster_args['num_clients'] < 2:
print("Not enough machines in the cluster to run the '%s' benchmark"
% name)
print("Need at least 2 machines in this configuration")
return
if options.numTables == None:
client_args['--numTables'] = 1
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), **cluster_args)
for i in range(1, cluster_args['num_clients'] + 1):
print(get_client_log(i), end='')
def transactionThroughput(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '-t 2000'
if cluster_args['timeout'] < 250:
cluster_args['timeout'] = 250
if 'num_clients' not in cluster_args:
cluster_args['num_clients'] = len(getHosts()) - cluster_args['num_servers']
if cluster_args['num_clients'] < 2:
print("Not enough machines in the cluster to run the '%s' benchmark"
% name)
print("Need at least 2 machines in this configuration")
return
if options.numTables == None:
client_args['--numTables'] = 1
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), **cluster_args)
for i in range(1, cluster_args['num_clients'] + 1):
print(get_client_log(i), end='')
def readDistRandom(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '-t 1000'
cluster.run(
client='%s/apps/ClusterPerf %s %s' %
(config.hooks.get_remote_obj_path(), flatten_args(client_args), name),
**cluster_args)
print(
"# Cumulative distribution of time for a single client to read a\n"
"# random %d-byte object from a single server. Each line indicates\n"
"# that a given fraction of all reads took at most a given time\n"
"# to complete.\n"
"# Generated by 'clusterperf.py readDist'\n#\n"
"# Time (usec) Cum. Fraction\n"
"#---------------------------" % options.size)
if (options.rcdf):
print_rcdf_from_log()
else:
print_cdf_from_log()
def writeDist(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '-t 2000'
cluster_args['disjunct'] = True
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name),
**cluster_args)
print(
"# Cumulative distribution of time for a single client to write a\n"
"# single %d-byte object from a single server. Each line indicates\n"
"# that a given fraction of all writes took at most a given time\n"
"# to complete.\n"
"# Generated by 'clusterperf.py %s'\n#\n"
"# Time (usec) Cum. Fraction\n"
"#---------------------------" % (options.size, name))
if (options.rcdf):
print_rcdf_from_log()
else:
print_cdf_from_log()
def indexRange(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '--maxCores 2 --totalMasterMemory 1500'
if cluster_args['timeout'] < 360:
cluster_args['timeout'] = 360
if '--numObjects' not in client_args:
client_args['--numObjects'] = 1000
if '--warmup' not in client_args:
client_args['--warmup'] = 10
if '--count' not in client_args:
client_args['--count'] = 90
# Ensure at least 5 hosts for optimal performance
if options.num_servers == None:
cluster_args['num_servers'] = len(getHosts())
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end='')
def readDistRandom(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '-t 1000'
cluster.run(client='%s/apps/ClusterPerf %s %s' %
(config.hooks.get_remote_obj_path(),
flatten_args(client_args), name),
**cluster_args)
print("# Cumulative distribution of time for a single client to read a\n"
"# random %d-byte object from a single server. Each line indicates\n"
"# that a given fraction of all reads took at most a given time\n"
"# to complete.\n"
"# Generated by 'clusterperf.py readDist'\n#\n"
"# Time (usec) Cum. Fraction\n"
"#---------------------------"
% options.size)
if (options.rcdf):
print_rcdf_from_log()
else:
print_cdf_from_log()
def default(
name, # Name of this test; passed through
# to ClusterPerf verbatim.
options, # The full set of command-line options.
cluster_args, # Proposed set of arguments to pass to
# cluster.run (extracted from options).
# Individual tests can override as
# appropriate for the test.
client_args, # Proposed set of arguments to pass to
# ClusterPerf (via cluster.run).
# Individual tests can override as
# needed for the test.
):
"""
This function is used as the invocation function for most tests;
it simply invokes ClusterPerf via cluster.run and prints the result.
"""
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end='')
def transactionDist(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '-t 2000'
if options.numTables == None:
client_args['--numTables'] = 1
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name),
**cluster_args)
print(
"# Cumulative distribution of time for a single client to commit a\n"
"# transactional read-write on a single %d-byte object from a\n"
"# single server. Each line indicates that a given fraction of all\n"
"# commits took at most a given time to complete.\n"
"# Generated by 'clusterperf.py %s'\n#\n"
"# Time (usec) Cum. Fraction\n"
"#---------------------------" % (options.size, name))
if (options.rcdf):
print_rcdf_from_log()
else:
print_cdf_from_log()
def default(
name, # Name of this test; passed through
# to ClusterPerf verbatim.
options, # The full set of command-line options.
cluster_args, # Proposed set of arguments to pass to
# cluster.run (extracted from options).
# Individual tests can override as
# appropriate for the test.
client_args, # Proposed set of arguments to pass to
# ClusterPerf (via cluster.run).
# Individual tests can override as
# needed for the test.
):
"""
This function is used as the invocation function for most tests;
it simply invokes ClusterPerf via cluster.run and prints the result.
"""
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end='')
def indexRange(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '--maxCores 2 --totalMasterMemory 1500'
if cluster_args['timeout'] < 360:
cluster_args['timeout'] = 360
if '--numObjects' not in client_args:
client_args['--numObjects'] = 1000
if '--warmup' not in client_args:
client_args['--warmup'] = 10
if '--count' not in client_args:
client_args['--count'] = 90
# Ensure at least 5 hosts for optimal performance
if options.num_servers == None:
cluster_args['num_servers'] = len(getHosts())
cluster.run(
client='%s/apps/ClusterPerf %s %s' %
(config.hooks.get_remote_obj_path(), flatten_args(client_args), name),
**cluster_args)
print(get_client_log(), end='')
def indexScalability(name, options, cluster_args, client_args):
if 'master_args' not in cluster_args:
cluster_args['master_args'] = '--maxCores 3'
if cluster_args['timeout'] < 360:
cluster_args['timeout'] = 360
cluster_args['backup_disks_per_server'] = 0
cluster_args['replicas'] = 0
# Number of concurrent rpcs to do per indexlet
if '--count' not in client_args:
client_args['--count'] = 20
# Number of objects per read request
if '--numObjects' not in client_args:
client_args['--numObjects'] = 1
# Ensure at least 15 hosts for optimal performance
if options.num_servers == None:
cluster_args['num_servers'] = len(getHosts())
if 'num_clients' not in cluster_args:
cluster_args['num_clients'] = 10
cluster.run(client='%s/ClusterPerf %s %s' %
(obj_path, flatten_args(client_args), name), **cluster_args)
print(get_client_log(), end='')