def runCachingExperiment(name, data_dir):
print("Running caching experiment for '" + name + "'.")
memoryPerWorker = "21g"
cores = 143
timesToRun = 4
tps = 1500000
outFilePath = data_dir + "/caching/" + name + ".yaml"
if os.path.isfile(outFilePath):
with open(outFilePath, "r") as f: data = yaml.load(f)
else: data = {}
for cache in [True,False]:
if cache not in data: data[cache] = []
if len(data[cache]) >= timesToRun:
print(" Already profiled for cache = " + str(cache) + ", skipping.")
continue
while len(data[cache]) < timesToRun:
try:
bench_base.restartServers()
bench_base.restartSparkContext(memoryPerWorker, cores)
# Load the data into cache.
if cache: bench_base.runQuery(name,"2014-01-01","2014-01-07",cache,tps)
while len(data[cache]) < timesToRun:
result = bench_base.runQuery(name,"2014-01-01","2014-01-07",cache,tps)
data[cache].append(result[2]['TimeMillis'] - result[0]['TimeMillis'])
with open(outFilePath, "w") as f:
f.write(yaml.dump(data, indent=2, default_flow_style=False))
except KeyboardInterrupt: sys.exit(-1)
except Exception:
print("Exception occurred, retrying.")
traceback.print_exc()
if not cache: data[cache] = []
pass
return data
def runScalingExperiment(queries, data_dir):
print("Running scaling experiment.")
memoryPerWorker = "21g"
timesToRun = 4
tps = 1500000
allNodes = [6,5,4,3]
cache = True
outFilePath = data_dir + "/scaling/scaling.yaml"
if os.path.isfile(outFilePath):
with open(outFilePath, "r") as f: data = yaml.load(f)
else: data = {}
for nodes in allNodes:
cores = nodes*24-1
if nodes not in data: data[nodes] = {'dataLoad': [], 'execution': {}}
if len(data[nodes]['dataLoad']) < timesToRun or \
len(data[nodes]['execution']) < len(queries):
input("Please ensure {} Spark and HDFS nodes are running and press any key to continue.".format(nodes))
while len(data[nodes]['dataLoad']) < timesToRun or \
len(data[nodes]['execution']) < len(queries):
try:
bench_base.restartServers()
bench_base.restartSparkContext(memoryPerWorker, cores)
while len(data[nodes]['dataLoad']) < timesToRun:
result = bench_base.getDataLoadTime()
data[nodes]['dataLoad'].append(
result[1]['TimeMillis'] - result[0]['TimeMillis']
)
with open(outFilePath, "w") as f:
f.write(yaml.dump(data, indent=2, default_flow_style=False))
for query in queries:
exeData = data[nodes]['execution']
if query not in exeData: exeData[query] = []
if len(exeData[query]) >= timesToRun:
print(" Already profiled for nodes = " + str(nodes) + ", skipping.")
continue
while len(exeData[query]) < timesToRun:
# Load the data into cache.
if cache:
bench_base.runQuery(query,"2014-01-01","2014-01-07",cache,tps)
while len(exeData[query]) < timesToRun:
result = bench_base.runQuery(
query,"2014-01-01","2014-01-07",cache,tps
)
exeData[query].append(
result[2]['TimeMillis'] - result[0]['TimeMillis']
)
with open(outFilePath, "w") as f:
f.write(yaml.dump(data, indent=2, default_flow_style=False))
except KeyboardInterrupt: sys.exit(-1)
except Exception:
print("Exception occurred, retrying.")
traceback.print_exc()
if not cache: data[cache] = []
pass
return data
def runPartitionExperiment(name, data_dir):
print("Running partition experiment for '" + name + "'.")
targetPartitionSizes = [
10000, 50000, 100000, 200000, 300000, 400000, 500000, 750000, 1000000,
1500000
]
memoryPerWorker = "21g"
cores = 143
timesToRun = 4
cache = True
outFilePath = data_dir + "/partitions/" + name + ".yaml"
if os.path.isfile(outFilePath):
with open(outFilePath, "r") as f:
data = yaml.load(f)
else:
data = {}
for tps in targetPartitionSizes:
if tps not in data: data[tps] = []
if len(data[tps]) >= timesToRun:
print(" Already profiled for " + str(tps) +
" targetPartitionSizes, skipping.")
continue
while len(data[tps]) < timesToRun:
try:
bench_base.restartServers()
bench_base.restartSparkContext(memoryPerWorker, cores)
# Load the data into cache.
if cache:
bench_base.runQuery(name, "2014-01-01", "2014-01-07",
cache, tps)
while len(data[tps]) < timesToRun:
result = bench_base.runQuery(name, "2014-01-01",
"2014-01-07", cache, tps)
data[tps].append(result[2]['TimeMillis'] -
result[0]['TimeMillis'])
with open(outFilePath, "w") as f:
f.write(yaml.dump(data, indent=2,
default_flow_style=False))
except KeyboardInterrupt:
sys.exit(-1)
except Exception:
print("Exception occurred, retrying.")
traceback.print_exc()
if not cache: data[tps] = []
pass
return data
def runPartitionExperiment(name, data_dir):
print("Running partition experiment for '" + name + "'.")
targetPartitionSizes = [10000, 50000, 100000, 200000,
300000, 400000, 500000, 750000, 1000000, 1500000]
memoryPerWorker = "21g"
cores = 143
timesToRun = 4
cache = True
outFilePath = data_dir + "/partitions/" + name + ".yaml"
if os.path.isfile(outFilePath):
with open(outFilePath, "r") as f: data = yaml.load(f)
else: data = {}
for tps in targetPartitionSizes:
if tps not in data: data[tps] = []
if len(data[tps]) >= timesToRun:
print(" Already profiled for " + str(tps) +
" targetPartitionSizes, skipping.")
continue
while len(data[tps]) < timesToRun:
try:
bench_base.restartServers()
bench_base.restartSparkContext(memoryPerWorker, cores)
# Load the data into cache.
if cache: bench_base.runQuery(name,"2014-01-01","2014-01-07",cache,tps)
while len(data[tps]) < timesToRun:
result = bench_base.runQuery(name,"2014-01-01","2014-01-07",cache,tps)
data[tps].append(result[2]['TimeMillis'] - result[0]['TimeMillis'])
with open(outFilePath, "w") as f:
f.write(yaml.dump(data, indent=2, default_flow_style=False))
except KeyboardInterrupt: sys.exit(-1)
except Exception:
print("Exception occurred, retrying.")
traceback.print_exc()
if not cache: data[tps] = []
pass
return data
def runScalingExperiment(queries, data_dir):
print("Running scaling experiment.")
memoryPerWorker = "21g"
timesToRun = 4
tps = 1500000
allNodes = [6, 5, 4, 3]
cache = True
outFilePath = data_dir + "/scaling/scaling.yaml"
if os.path.isfile(outFilePath):
with open(outFilePath, "r") as f:
data = yaml.load(f)
else:
data = {}
for nodes in allNodes:
cores = nodes * 24 - 1
if nodes not in data: data[nodes] = {'dataLoad': [], 'execution': {}}
if len(data[nodes]['dataLoad']) < timesToRun or \
len(data[nodes]['execution']) < len(queries):
input(
"Please ensure {} Spark and HDFS nodes are running and press any key to continue."
.format(nodes))
while len(data[nodes]['dataLoad']) < timesToRun or \
len(data[nodes]['execution']) < len(queries):
try:
bench_base.restartServers()
bench_base.restartSparkContext(memoryPerWorker, cores)
while len(data[nodes]['dataLoad']) < timesToRun:
result = bench_base.getDataLoadTime()
data[nodes]['dataLoad'].append(result[1]['TimeMillis'] -
result[0]['TimeMillis'])
with open(outFilePath, "w") as f:
f.write(yaml.dump(data, indent=2,
default_flow_style=False))
for query in queries:
exeData = data[nodes]['execution']
if query not in exeData: exeData[query] = []
if len(exeData[query]) >= timesToRun:
print(" Already profiled for nodes = " + str(nodes) +
", skipping.")
continue
while len(exeData[query]) < timesToRun:
# Load the data into cache.
if cache:
bench_base.runQuery(query, "2014-01-01",
"2014-01-07", cache, tps)
while len(exeData[query]) < timesToRun:
result = bench_base.runQuery(
query, "2014-01-01", "2014-01-07", cache, tps)
exeData[query].append(result[2]['TimeMillis'] -
result[0]['TimeMillis'])
with open(outFilePath, "w") as f:
f.write(
yaml.dump(data,
indent=2,
default_flow_style=False))
except KeyboardInterrupt:
sys.exit(-1)
except Exception:
print("Exception occurred, retrying.")
traceback.print_exc()
if not cache: data[cache] = []
pass
return data
def runConcurrentExperiment(name, data_dir):
global timesToRun
print("Running concurrent experiment for '" + name + "'.")
allConcurrentQueries = list(range(1,9))
cores = 143
memoryPerWorker = "20g"
tps = 1500000
def isDataFinished(concurrentQueries,d):
if not d or not isinstance(d,list) or len(d)<concurrentQueries:
return False
for thread in d:
if len(thread) < timesToRun: return False
return True
def run(concurrentQueries):
g_lock = Lock()
def threadEntry(threadNum):
def isFinished():
with g_lock:
for n in results:
if len(n) < timesToRun: return False
return True
try:
while not isFinished():
print(str(threadNum) + ": Calling query.")
result = bench_base.runQuery(name, "2014-01-01", "2014-01-07", True)
queryExecutionTime = result[2]['TimeMillis']-result[0]['TimeMillis']
print(str(threadNum) + ": Query execution time: " +
str(queryExecutionTime))
with g_lock:
results[threadNum].append(queryExecutionTime)
except:
print("Error occurred in thread.")
traceback.print_exc()
results = [[] for x in range(0,concurrentQueries)]
threads = [
Thread(target=threadEntry, args=(i,)) for i in range(0,concurrentQueries)
]
[t.start() for t in threads]
[t.join() for t in threads]
return results
outFilePath = data_dir + "/concurrent/" + name + ".yaml"
if os.path.isfile(outFilePath):
with open(outFilePath, "r") as f: data = yaml.load(f)
else: data = {}
for concurrentQueries in allConcurrentQueries:
if concurrentQueries in data and \
isDataFinished(concurrentQueries,data[concurrentQueries]):
print(" Already profiled for " + str(concurrentQueries) +
" concurrent queries, skipping.")
continue
else:
data[concurrentQueries] = {}
while not isDataFinished(concurrentQueries,data[concurrentQueries]):
try:
bench_base.restartServers()
bench_base.restartSparkContext(memoryPerWorker, cores)
# For cache.
bench_base.runQuery(name, "2014-01-01", "2014-01-07", True, tps)
data[concurrentQueries] = run(concurrentQueries)
with open(outFilePath, "w") as f:
f.write(yaml.dump(data, indent=2, default_flow_style=False))
except KeyboardInterrupt: sys.exit(-1)
except Exception:
print("Exception occurred, retrying.")
traceback.print_exc()
data[concurrentQueries] = {}
pass
return data
def runConcurrentExperiment(name, data_dir):
global timesToRun
print("Running concurrent experiment for '" + name + "'.")
allConcurrentQueries = list(range(1, 9))
cores = 143
memoryPerWorker = "20g"
tps = 1500000
def isDataFinished(concurrentQueries, d):
if not d or not isinstance(d, list) or len(d) < concurrentQueries:
return False
for thread in d:
if len(thread) < timesToRun: return False
return True
def run(concurrentQueries):
g_lock = Lock()
def threadEntry(threadNum):
def isFinished():
with g_lock:
for n in results:
if len(n) < timesToRun: return False
return True
try:
while not isFinished():
print(str(threadNum) + ": Calling query.")
result = bench_base.runQuery(name, "2014-01-01",
"2014-01-07", True)
queryExecutionTime = result[2]['TimeMillis'] - result[0][
'TimeMillis']
print(
str(threadNum) + ": Query execution time: " +
str(queryExecutionTime))
with g_lock:
results[threadNum].append(queryExecutionTime)
except:
print("Error occurred in thread.")
traceback.print_exc()
results = [[] for x in range(0, concurrentQueries)]
threads = [
Thread(target=threadEntry, args=(i, ))
for i in range(0, concurrentQueries)
]
[t.start() for t in threads]
[t.join() for t in threads]
return results
outFilePath = data_dir + "/concurrent/" + name + ".yaml"
if os.path.isfile(outFilePath):
with open(outFilePath, "r") as f:
data = yaml.load(f)
else:
data = {}
for concurrentQueries in allConcurrentQueries:
if concurrentQueries in data and \
isDataFinished(concurrentQueries,data[concurrentQueries]):
print(" Already profiled for " + str(concurrentQueries) +
" concurrent queries, skipping.")
continue
else:
data[concurrentQueries] = {}
while not isDataFinished(concurrentQueries, data[concurrentQueries]):
try:
bench_base.restartServers()
bench_base.restartSparkContext(memoryPerWorker, cores)
# For cache.
bench_base.runQuery(name, "2014-01-01", "2014-01-07", True,
tps)
data[concurrentQueries] = run(concurrentQueries)
with open(outFilePath, "w") as f:
f.write(yaml.dump(data, indent=2,
default_flow_style=False))
except KeyboardInterrupt:
sys.exit(-1)
except Exception:
print("Exception occurred, retrying.")
traceback.print_exc()
data[concurrentQueries] = {}
pass
return data