def benchmark(loopcount, workers, matn, verbose=False):
t1 = time.time()
N = workers
loopcounts = [(loopcount,N),]*N
source = pyLambdaFlows.Source()
compute_op = pyLambdaFlows.Map(source, "./compute.py")
sess = pyLambdaFlows.Session(credentials_csv="./accessKeys.csv")
compute_op.compile(sess=sess, purge=True)
promess = compute_op.eval(feed_dict={source: loopcounts}, wait=False)
print("invocation done, dur=", time.time() - t1)
local_jobs_done_timeline = []
result_count = 0
while result_count < N:
result_count = promess.getStatus()
local_jobs_done_timeline.append((time.time(), result_count))
est_flop = 2 * result_count * loopcount * matn ** 3
est_gflops = est_flop / 1e9 / (time.time() - t1)
if verbose:
print("jobs done: {:5d} runtime: {:5.1f}s {:8.1f} GFLOPS ".format(result_count,
time.time() - t1,
est_gflops))
if result_count == N:
break
time.sleep(1)
if verbose:
print("getting results")
#results = [f.result(throw_except=False) for f in futures]
results = promess.getResult()
if verbose:
print("getting status")
run_statuses = [f.run_status for f in futures]
invoke_statuses = [f.invoke_status for f in futures]
all_done = time.time()
total_time = all_done - t1
print("total time", total_time)
est_flop = result_count * 2 * loopcount * matn ** 3
print(est_flop / 1e9 / total_time, "GFLOPS")
res = {'total_time': total_time,
'est_flop': est_flop,
'run_statuses': run_statuses,
'invoke_statuses': invoke_statuses,
'callset_id': futures[0].callset_id,
'local_jobs_done_timeline': local_jobs_done_timeline,
'results': results}
return res
def run():
start=time()
# Creation des sources pyLambdaFlow, cad, l'origine des invocations, on lui donne les donnees d'entrees a la ligne 48
param = pyLambdaFlows.Source()
param.files.append("./population.py") # Il est important de passer a la source les fichiers necessaires pour l'execution des autres fonctions Lambda
param.files.append("./schedule.py")
param.files.append("./domain.py")
param.files.append("./utils.py")
param.files.append("./data.py")
param.files.append("./genetic_algorithm.py")
# Le kernel est appelle par un operateur Map et la liste d'entree (param_list) comporte SPLIT_VAR element, le kernel Create_&_Evaluate sera donc appelle sur chaque element de param_list
# Il faut lui passer en donnees la liste des fichiers necessaires a son execution
b = pyLambdaFlows.op.Map(param, ["./Create_And_Evaluate.py", "./population.py", "./schedule.py", "./domain.py", "./utils.py", "./data.py", "./genetic_algorithm.py"]) # dependances size=SPLIT_VAR : [[0], [1], [2], [3]]
# Les kernels suivant utilisent un operateur cree specialement pour realiser un Map mais pour que tous les resultats soient envoyes a tous les kernels de la layer suivante (voir schema du rapport)
c = MapToMapOp([b, param], ["./Rate_And_Generate.py", "./population.py", "./schedule.py", "./domain.py", "./utils.py", "./data.py", "./genetic_algorithm.py"]) # dependances size = SPLIT_VAR : [[0, 1, 2, 3], [0, 1, 2, 3], [0, 1, 2, 3], [0, 1, 2, 3]]
d = MapToMapOp([c, param], ["./Rate_And_Generate.py", "./population.py", "./schedule.py", "./domain.py", "./utils.py", "./data.py", "./genetic_algorithm.py"])
e = MapToMapOp([d, param], ["./Rate_And_Generate.py", "./population.py", "./schedule.py", "./domain.py", "./utils.py", "./data.py", "./genetic_algorithm.py"])
f = MapToMapOp([e, param], ["./Rate_And_Generate.py", "./population.py", "./schedule.py", "./domain.py", "./utils.py", "./data.py", "./genetic_algorithm.py"])
# Creation de la liste de parametre envoyee en entree de la source
param_list = []
for _ in range (SPLIT_VAR):
param_list.append([POPULATION_SIZE,MUTATION_RATE,CROSSOVER_RATE,TOURNAMENT_SELECTION_SIZE,SPLIT_VAR,NUMB_OF_ELITE_SCHEDULES,NUMB_OF_GENERATION])
# Creation d'une nouvelle session Lambda avec les informations de connexion dans 'accessKeys.csv', et lancement de l'invocation des Kernels
with pyLambdaFlows.Session(credentials_csv="./accessKeys.csv") as sess:
f.compile(purge=False)
result = f.eval(feed_dict={param:param_list})
end=time()
# Visualisation du resultat
#print(result)
res_list=[]
for i in range (SPLIT_VAR):
for s in range(NUMB_OF_ELITE_SCHEDULES):
#print(result[i][s]._fitness)
res_list.append(result[i][s])
print("Temps d'exécution : {}".format(end-start))
res = sorted(res_list,key = lambda schedule: schedule._fitness,reverse=True)
print("Schedule: {} avec une fitness = {}".format(res[0],res[0]._fitness))
final_res = (res[0]._fitness,(end-start))
return final_res
def run():
start = time()
# Creation des sources pyLambdaFlow, cad, l'origine des invocations, on lui donne les donnees d'entrees a la ligne 41
param = pyLambdaFlows.Source()
param.files.append(
"./population.py"
) # Il est important de passer a la source les fichiers necessaires pour l'execution des autres fonctions Lambda
param.files.append("./schedule.py")
param.files.append("./domain.py")
param.files.append("./utils.py")
param.files.append("./data.py")
param.files.append("./genetic_algorithm.py")
# Le kernel est appelle par un operateur Map mais la liste d'entree (param_list) ne comportant qu'un seul element, ce mapper ne sera appelle qu'une fois
# Il faut lui passer en donnees la liste des fichiers necessaires a son execution
b = pyLambdaFlows.op.Map(param, [
"./Sequential_Kernel.py", "./population.py", "./schedule.py",
"./domain.py", "./utils.py", "./data.py", "./genetic_algorithm.py"
]) # dependances avec size=SPLIT_VAR = [[0]]
# Creation de la liste de parametre envoyee en entree de la source
param_list = []
for _ in range(SPLIT_VAR):
param_list.append([
POPULATION_SIZE, MUTATION_RATE, CROSSOVER_RATE,
TOURNAMENT_SELECTION_SIZE, SPLIT_VAR, NUMB_OF_ELITE_SCHEDULES,
NUMB_OF_GENERATION
])
# Creation d'une nouvelle session Lambda avec les informations de connexion dans 'accessKeys.csv', et lancement de l'invocation des Kernels
with pyLambdaFlows.Session(credentials_csv="./accessKeys.csv") as sess:
b.compile(purge=False)
result = b.eval(feed_dict={param: param_list})
end = time()
# Visualisation du resultat
print(result)
print("Temps d'exécution : {}".format(end - start))
print("Schedule: {} avec une fitness = {}".format(result[0],
result[0]._fitness))
final_res = (result[0]._fitness, (end - start))
return final_res
import pyLambdaFlows
sess = pyLambdaFlows.Session(credentials_csv="./accessKeys.csv")
#a = pyLambdaFlows.upload.Uploader(sess)
#a.upload_lambda("./source/mean.py")
a = sess.getLambda()
elements = a.list_functions()["Functions"]
for element in elements:
a.delete_function(FunctionName=element["FunctionName"])
import pyLambdaFlows
from random import randint
a = pyLambdaFlows.op.Source()
b = pyLambdaFlows.op.Source()
c = pyLambdaFlows.op.Map([a, b], "./source/addition.py")
d = pyLambdaFlows.op.Reduce(c, "./source/mean.py")
json = None
with pyLambdaFlows.Session(credentials_csv="./accessKeys.csv",
auto_purge=True) as sess:
d.compile(purge=False)
result = d.eval(feed_dict={a: [1, 2, 1], b: [1, 2, 3]})
print(result)
