Improved handling of metadata, extended benchmark launch and templates to multinode benchmarks
This commit is contained in:
parent
ba8cb1ae01
commit
a25f8ffec6
127
launch_bench.py
Executable file
127
launch_bench.py
Executable file
@ -0,0 +1,127 @@
|
||||
import os
|
||||
import subprocess
|
||||
from datetime import datetime
|
||||
|
||||
################ HELPER FUNCTIONS ################
|
||||
|
||||
|
||||
def load_template(template_path: str):
|
||||
output_template = ""
|
||||
with open(template_path, "r") as handle:
|
||||
output_template = handle.read()
|
||||
return output_template
|
||||
|
||||
|
||||
def write_batch(batch_fpath: str, batch_content: str):
|
||||
with open(batch_fpath, "w") as handle:
|
||||
_ = handle.write(batch_content)
|
||||
|
||||
|
||||
################### SETUP DIRS ###################
|
||||
output_dir = os.getcwd()+"/output/"
|
||||
err_dir = os.getcwd()+"/error/"
|
||||
batch_files_dir = os.getcwd()+"/batchs/"
|
||||
data_dir = os.getcwd()+"/data/"
|
||||
|
||||
if os.path.isdir(output_dir) == False:
|
||||
os.mkdir(output_dir)
|
||||
if os.path.isdir(err_dir) == False:
|
||||
os.mkdir(err_dir)
|
||||
if os.path.isdir(data_dir) == False:
|
||||
os.mkdir(data_dir)
|
||||
if os.path.isdir(batch_files_dir) == False:
|
||||
os.mkdir(batch_files_dir)
|
||||
|
||||
################ GLOBAL DEFAULTS #################
|
||||
mpi1_bin = "/home/hpc/ihpc/ihpc136h/workspace/prototyping/bin/IMB-MPI1"
|
||||
default_parameter = {
|
||||
"time_stamp": datetime.now().strftime("%y_%m_%d_%H-%M-%S"),
|
||||
"job_name": "",
|
||||
"output_dir": os.getcwd()+"/output/",
|
||||
"err_dir": os.getcwd()+"/error/",
|
||||
"data_dir": os.getcwd()+"/data/",
|
||||
"n_procs": 18,
|
||||
"off_cache_flag": "",
|
||||
"bin": mpi1_bin,
|
||||
"n_nodes": 1
|
||||
}
|
||||
|
||||
collectives = [
|
||||
"Reduce",
|
||||
"Reduce_scatter",
|
||||
"Allreduce",
|
||||
"Allgather",
|
||||
"Allgatherv",
|
||||
"Scatter",
|
||||
"Scatterv",
|
||||
"Gather",
|
||||
"Gatherv",
|
||||
"Alltoall",
|
||||
"Bcast",
|
||||
]
|
||||
|
||||
log = ""
|
||||
|
||||
############### SINGLE-NODE LAUNCH ###############
|
||||
procnt = [
|
||||
18,
|
||||
36,
|
||||
54,
|
||||
72
|
||||
]
|
||||
|
||||
off_cache_flags = [
|
||||
"-off_cache -1",
|
||||
"-off_cache 50",
|
||||
""
|
||||
]
|
||||
|
||||
single_node_parameter = dict(default_parameter)
|
||||
single_node_template = load_template("templates/singlenode.template")
|
||||
|
||||
for flag in off_cache_flags:
|
||||
single_node_parameter["off_cache_flag"] = flag
|
||||
for n_procs in procnt:
|
||||
single_node_parameter["n_procs"] = n_procs
|
||||
for collective in collectives:
|
||||
single_node_parameter["job_name"] = collective
|
||||
write_batch(batch_files_dir+collective+".sh",
|
||||
single_node_template.format(**single_node_parameter))
|
||||
result = subprocess.run(["sbatch", batch_files_dir+collective+".sh"],
|
||||
capture_output=True, text=True)
|
||||
log += f"#{collective} {n_procs}" + "\n"
|
||||
log += "\tSTDOUT:" + result.stdout + "\n"
|
||||
log += "\tSTDERR:" + result.stderr + "\n"
|
||||
############## MULTIPLE-NODE LAUNCH ##############
|
||||
off_cache_flags = [
|
||||
"-off_cache -1",
|
||||
"-off_cache 50",
|
||||
""
|
||||
]
|
||||
|
||||
ndcnt = [
|
||||
2,
|
||||
3,
|
||||
4
|
||||
]
|
||||
|
||||
proc_per_node = 72
|
||||
multiple_node_parameter = dict(default_parameter)
|
||||
multiple_node_template = load_template("templates/multinode.template")
|
||||
|
||||
for flag in off_cache_flags:
|
||||
multiple_node_parameter["off_cache_flag"] = flag
|
||||
for n_nodes in ndcnt:
|
||||
n_procs = n_nodes*proc_per_node
|
||||
multiple_node_parameter["n_procs"] = int(n_procs)
|
||||
multiple_node_parameter["n_nodes"] = n_nodes
|
||||
for collective in collectives:
|
||||
multiple_node_parameter["job_name"] = collective
|
||||
write_batch(batch_files_dir+collective+".sh",
|
||||
multiple_node_template.format(**multiple_node_parameter))
|
||||
result = subprocess.run(["sbatch", batch_files_dir+collective+".sh"],
|
||||
capture_output=True, text=True)
|
||||
log += f"#{collective} {n_procs}" + "\n"
|
||||
log += "\tSTDOUT:" + result.stdout + "\n"
|
||||
log += "\tSTDERR:" + result.stderr + "\n"
|
||||
print(log)
|
||||
@ -1,79 +0,0 @@
|
||||
import os
|
||||
import subprocess
|
||||
from datetime import datetime
|
||||
|
||||
|
||||
def load_template(template_path: str):
|
||||
output_template = ""
|
||||
with open(template_path, "r") as handle:
|
||||
output_template = handle.read()
|
||||
return output_template
|
||||
|
||||
|
||||
def write_batch(batch_fpath: str, batch_content: str):
|
||||
with open(batch_fpath, "w") as handle:
|
||||
_ = handle.write(batch_content)
|
||||
|
||||
|
||||
collectives = ["Reduce",
|
||||
# "Reduce_scatter",
|
||||
# "Reduce_scatter_block",
|
||||
# "Allreduce",
|
||||
# "Allgather",
|
||||
# "Allgatherv",
|
||||
# "Scatter",
|
||||
# "Scatterv",
|
||||
# "Gather",
|
||||
# "Gatherv",
|
||||
# "Alltoall",
|
||||
# "Bcast",
|
||||
# "Barrier"
|
||||
]
|
||||
|
||||
procnt = [
|
||||
18,
|
||||
# 36,
|
||||
# 54,
|
||||
# 72
|
||||
]
|
||||
mpi1_bin = "/home/hpc/ihpc/ihpc136h/workspace/prototyping/bin"
|
||||
slurm_template = load_template("templates/bench.template")
|
||||
|
||||
template_parameter = {"time_stamp": datetime.now().strftime("%y_%m_%d_%H-%M-%S"),
|
||||
"job_name": "",
|
||||
"output_dir": os.getcwd()+"/output/",
|
||||
"err_dir": os.getcwd()+"/error/",
|
||||
"data_dir": os.getcwd()+"/data/",
|
||||
"n_procs": 18,
|
||||
"off_mem_flag": "",
|
||||
"bin": mpi1_bin
|
||||
}
|
||||
|
||||
output_dir = os.getcwd()+"/output/"
|
||||
err_dir = os.getcwd()+"/error/"
|
||||
batch_files_dir = os.getcwd()+"/batchs/"
|
||||
data_dir = os.getcwd()+"/data/"
|
||||
|
||||
if os.path.isdir(output_dir) == False:
|
||||
os.mkdir(output_dir)
|
||||
if os.path.isdir(err_dir) == False:
|
||||
os.mkdir(err_dir)
|
||||
if os.path.isdir(data_dir) == False:
|
||||
os.mkdir(data_dir)
|
||||
if os.path.isdir(batch_files_dir) == False:
|
||||
os.mkdir(batch_files_dir)
|
||||
|
||||
log = ""
|
||||
|
||||
for n_procs in procnt:
|
||||
template_parameter["n_procs"] = n_procs
|
||||
for collective in collectives:
|
||||
template_parameter["job_name"] = collective
|
||||
write_batch(batch_files_dir+collective+".sh",
|
||||
slurm_template.format(**template_parameter))
|
||||
result = subprocess.run(["sbatch", batch_files_dir+collective+".sh"],
|
||||
capture_output=True, text=True)
|
||||
log += f"#{collective} {n_procs}" + "\n"
|
||||
log += "\tSTDOUT:" + result.stdout + "\n"
|
||||
log += "\tSTDERR:" + result.stderr + "\n"
|
||||
print(log)
|
||||
@ -1,83 +0,0 @@
|
||||
import os
|
||||
import subprocess
|
||||
from datetime import datetime
|
||||
|
||||
|
||||
def load_template(template_path: str):
|
||||
output_template = ""
|
||||
with open(template_path, "r") as handle:
|
||||
output_template = handle.read()
|
||||
return output_template
|
||||
|
||||
|
||||
def write_batch(batch_fpath: str, batch_content: str):
|
||||
with open(batch_fpath, "w") as handle:
|
||||
_ = handle.write(batch_content)
|
||||
|
||||
|
||||
collectives = [
|
||||
"Reduce",
|
||||
"Reduce_scatter",
|
||||
"Allreduce",
|
||||
"Allgather",
|
||||
"Allgatherv",
|
||||
"Scatter",
|
||||
"Scatterv",
|
||||
"Gather",
|
||||
"Gatherv",
|
||||
"Alltoall",
|
||||
"Bcast",
|
||||
# "Barrier"
|
||||
]
|
||||
|
||||
procnt = [
|
||||
18,
|
||||
36,
|
||||
54,
|
||||
72
|
||||
]
|
||||
|
||||
mpi1_bin = "/home/hpc/ihpc/ihpc136h/workspace/prototyping/bin/IMB-MPI1"
|
||||
slurm_template = load_template("templates/bench.template")
|
||||
|
||||
template_parameter = {
|
||||
"time_stamp": datetime.now().strftime("%y_%m_%d_%H-%M-%S"),
|
||||
"job_name": "",
|
||||
"output_dir": os.getcwd()+"/output/",
|
||||
"err_dir": os.getcwd()+"/error/",
|
||||
"data_dir": os.getcwd()+"/data/",
|
||||
"n_procs": 18,
|
||||
"off_mem_flag": "-off_cache 50",
|
||||
"bin": mpi1_bin,
|
||||
"n_nodes": 1
|
||||
}
|
||||
|
||||
output_dir = os.getcwd()+"/output/"
|
||||
err_dir = os.getcwd()+"/error/"
|
||||
batch_files_dir = os.getcwd()+"/batchs/"
|
||||
data_dir = os.getcwd()+"/data/"
|
||||
|
||||
if os.path.isdir(output_dir) == False:
|
||||
os.mkdir(output_dir)
|
||||
if os.path.isdir(err_dir) == False:
|
||||
os.mkdir(err_dir)
|
||||
if os.path.isdir(data_dir) == False:
|
||||
os.mkdir(data_dir)
|
||||
if os.path.isdir(batch_files_dir) == False:
|
||||
os.mkdir(batch_files_dir)
|
||||
|
||||
log = ""
|
||||
|
||||
for n_procs in procnt:
|
||||
template_parameter["n_procs"] = n_procs
|
||||
for collective in collectives:
|
||||
template_parameter["job_name"] = collective
|
||||
write_batch(batch_files_dir+collective+".sh",
|
||||
slurm_template.format(**template_parameter))
|
||||
result = subprocess.run(["sbatch", batch_files_dir+collective+".sh"],
|
||||
capture_output=True, text=True)
|
||||
log += f"#{collective} {n_procs}" + "\n"
|
||||
log += "\tSTDOUT:" + result.stdout + "\n"
|
||||
log += "\tSTDERR:" + result.stderr + "\n"
|
||||
print(log)
|
||||
# _ = subprocess.run(["./clean.sh"])
|
||||
@ -14,7 +14,7 @@ data_markers = {
|
||||
"end_of_table": "# All processes entering MPI_Finalize",
|
||||
"creation_time": "# CREATION_TIME :",
|
||||
"n_nodes": "# N_NODES :",
|
||||
"off_mem_flag": "# OFF_MEM_FLAG :"
|
||||
"off_cache_flag": "# OFF_CACHE_FLAG :"
|
||||
}
|
||||
|
||||
column_names = [
|
||||
@ -30,7 +30,7 @@ column_names = [
|
||||
"mpi_red_op",
|
||||
"creation_time",
|
||||
"n_nodes",
|
||||
"off_mem_flag",
|
||||
"off_cache_flag",
|
||||
]
|
||||
|
||||
data = list()
|
||||
@ -50,7 +50,7 @@ for file in os.listdir("data/"):
|
||||
mpi_red_op = "NA"
|
||||
creation_time = "NA"
|
||||
n_nodes = "NA"
|
||||
off_mem_flag = "NA"
|
||||
off_cache_flag = "NA"
|
||||
|
||||
for line in lines:
|
||||
if data_markers["block_separator"] in line:
|
||||
@ -73,10 +73,10 @@ for file in os.listdir("data/"):
|
||||
n_nodes = line.split()[-1]
|
||||
if data_markers["creation_time"] in line:
|
||||
creation_time = line.split()[-1]
|
||||
if data_markers["off_mem_flag"] in line:
|
||||
off_mem_flag = line.split(":")[-1].strip()
|
||||
if off_mem_flag == "": off_mem_flag = "NA"
|
||||
else: off_mem_flag = off_mem_flag.replace("-off_cache","")
|
||||
if data_markers["off_cache_flag"] in line:
|
||||
off_cache_flag = line.split(":")[-1].strip()
|
||||
if off_cache_flag == "": off_cache_flag = "NA"
|
||||
else: off_cache_flag = off_cache_flag.replace("-off_cache","")
|
||||
|
||||
if past_preheader and in_header:
|
||||
if data_markers["benchmark_type"] in line:
|
||||
@ -96,7 +96,7 @@ for file in os.listdir("data/"):
|
||||
mpi_red_op,
|
||||
creation_time,
|
||||
n_nodes,
|
||||
off_mem_flag,
|
||||
off_cache_flag,
|
||||
])
|
||||
|
||||
df = pd.DataFrame(data, columns=column_names)
|
||||
|
||||
22
templates/multinode.template
Normal file
22
templates/multinode.template
Normal file
@ -0,0 +1,22 @@
|
||||
#!/bin/bash -l
|
||||
#SBATCH --job-name={job_name}_{n_procs}
|
||||
#SBATCH --output={output_dir}{job_name}_{n_procs}.out
|
||||
#SBATCH --error={err_dir}{job_name}_{n_procs}.err
|
||||
#SBATCH --nodes={n_nodes}
|
||||
#SBATCH --time=00:10:00
|
||||
#SBATCH --export=NONE
|
||||
|
||||
unset SLURM_EXPORT_ENV
|
||||
|
||||
module load intel intelmpi
|
||||
|
||||
|
||||
OUTPUT_FILENAME="{data_dir}/{job_name}_$SLURM_JOB_ID.dat"
|
||||
|
||||
echo "# CREATION_TIME : {time_stamp}" > $OUTPUT_FILENAME
|
||||
echo "# N_NODES : {n_nodes}" >> $OUTPUT_FILENAME
|
||||
echo "# OFF_CACHE_FLAG : {off_cache_flag}">> $OUTPUT_FILENAME
|
||||
|
||||
srun --cpu-freq=2000000-2000000:performance -N {n_nodes} -n{n_procs} {bin} {job_name} -npmin {n_procs} {off_cache_flag} >> $OUTPUT_FILENAME
|
||||
|
||||
|
||||
@ -17,8 +17,8 @@ OUTPUT_FILENAME="{data_dir}/{job_name}_$SLURM_JOB_ID.dat"
|
||||
|
||||
echo "# CREATION_TIME : {time_stamp}" > $OUTPUT_FILENAME
|
||||
echo "# N_NODES : {n_nodes}" >> $OUTPUT_FILENAME
|
||||
echo "# OFF_MEM_FLAG : {off_mem_flag}">> $OUTPUT_FILENAME
|
||||
echo "# OFF_CACHE_FLAG : {off_cache_flag}">> $OUTPUT_FILENAME
|
||||
|
||||
srun --cpu-freq=2000000-2000000:performance ./likwid-mpirun -np {n_procs} -mpi intelmpi -omp intel -nperdomain M:18 {bin} {job_name} -npmin {n_procs} {off_mem_flag} >> $OUTPUT_FILENAME
|
||||
srun --cpu-freq=2000000-2000000:performance ./likwid-mpirun -np {n_procs} -mpi intelmpi -omp intel -nperdomain M:18 {bin} {job_name} -npmin {n_procs} {off_cache_flag} >> $OUTPUT_FILENAME
|
||||
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user