/
example_performanceProfiling.py
288 lines (254 loc) · 11.4 KB
/
example_performanceProfiling.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
from parcels import FieldSet, ParticleSet, JITParticle, AdvectionRK4
from datetime import timedelta as delta
from argparse import ArgumentParser
import numpy as np
import dask as da
import dask.array as daArray
from glob import glob
import time as ostime
import matplotlib.pyplot as plt
import os
import parcels
import psutil
import gc
import sys
try:
from mpi4py import MPI
except:
MPI = None
with_GC = False
with_ChunkInfoPrint = False
global_t_0 = 0
global_m_0 = 0
global_samples = []
samplenr = 0
global_memory_step = []
global_time_steps = []
global_fds_step = []
class IterationCounter():
_iter = 0
@classmethod
def advance(self):
old_iter = self._iter
self._iter += 1
return old_iter
class PerformanceLog():
samples = []
times_steps = []
memory_steps = []
fds_steps = []
_iter = 0
def advance(self):
if MPI:
mpi_comm = MPI.COMM_WORLD
mpi_rank = mpi_comm.Get_rank()
process = psutil.Process(os.getpid())
mem_B_used = process.memory_info().rss
fds_open = len(process.open_files())
mem_B_used_total = mpi_comm.reduce(mem_B_used, op=MPI.SUM, root=0)
fds_open_total = mpi_comm.reduce(fds_open, op=MPI.SUM, root=0)
if mpi_rank == 0:
self.times_steps.append(ostime.time())
self.memory_steps.append(mem_B_used_total)
self.fds_steps.append(fds_open_total)
self.samples.append(self._iter)
self._iter += 1
else:
process = psutil.Process(os.getpid())
self.times_steps.append(ostime.time())
self.memory_steps.append(process.memory_info().rss)
self.fds_steps.append(len(process.open_files()))
self.samples.append(self._iter)
self._iter += 1
def set_cmems_fieldset(cs, deferLoadFlag=True, periodicFlag=False):
ddir_head = "/data/oceanparcels/input_data"
ddir = os.path.join(ddir_head, "CMEMS/GLOBAL_REANALYSIS_PHY_001_030/")
files = sorted(glob(ddir+"mercatorglorys12v1_gl12_mean_201607*.nc"))
variables = {'U': 'uo', 'V': 'vo'}
dimensions = {'lon': 'longitude', 'lat': 'latitude', 'time': 'time'}
if cs not in ['auto', False]:
# cs = (1, cs, cs) # == OLD initialization as tuple. Advised to rather use a dictionary. ==#
cs = {dimensions['time']: 1, dimensions['lon']: cs, dimensions['lat']: cs}
if periodicFlag:
return FieldSet.from_netcdf(files, variables, dimensions, time_periodic=delta(days=30), deferred_load=deferLoadFlag, field_chunksize=cs)
else:
return FieldSet.from_netcdf(files, variables, dimensions, allow_time_extrapolation=True, deferred_load=deferLoadFlag, field_chunksize=cs)
def print_field_info(fieldset):
for f in fieldset.get_fields():
if type(f) in [parcels.VectorField, parcels.NestedField, parcels.SummedField] or not f.grid.defer_load:
continue
if isinstance(f.data, daArray.core.Array):
sys.stdout.write("Array of Field[name={}] is dask.Array\n".format(f.name))
sys.stdout.write(
"Chunk info of Field[name={}]: field.nchunks={}; shape(field.data.nchunks)={}; field.data.numblocks={}; shape(f.data)={}\n".format(
f.name, f.nchunks, (len(f.data.chunks[0]), len(f.data.chunks[1]), len(f.data.chunks[2])),
f.data.numblocks, f.data.shape))
sys.stdout.write("Chunk info of Grid[field.name={}]: g.chunk_info={}; g.load_chunk={}; len(g.load_chunk)={}\n".format(
f.name, f.grid.chunk_info, f.grid.load_chunk, len(f.grid.load_chunk)))
def plot(x, times, memory_used, nfiledescriptors, imageFilePath):
plot_t = []
for i in range(len(times)):
if i == 0:
plot_t.append(times[i]-global_t_0)
else:
plot_t.append(times[i]-times[i-1])
mem_scaler = (1*10)/(1024*1024*1024)
plot_mem = []
for i in range(len(memory_used)):
plot_mem.append(memory_used[i] * mem_scaler)
fig, ax = plt.subplots(1, 1, figsize=(15, 12))
ax.plot(x, plot_t, 'o-', label="time_spent [s]")
ax.plot(x, plot_mem, 'x-', label="memory_used [100 MB]")
ax.plot(x, nfiledescriptors, '.-', label="open_files [#]")
plt.ylim([0, 50])
plt.legend()
ax.set_xlabel('iteration')
plt.savefig(os.path.join(odir, imageFilePath), dpi=300, format='png')
def LogMemTimeFds():
if MPI:
mpi_comm = MPI.COMM_WORLD
mpi_rank = mpi_comm.Get_rank()
process = psutil.Process(os.getpid())
mem_B_used = process.memory_info().rss
fds_open = len(psutil.Process().open_files())
mem_B_used_total = mpi_comm.reduce(mem_B_used, op=MPI.SUM, root=0)
fds_open_total = mpi_comm.reduce(fds_open, op=MPI.SUM, root=0)
if mpi_rank == 0:
global_time_steps.append(ostime.time())
global_memory_step.append(mem_B_used_total)
global_fds_step.append(fds_open_total)
global_samples.append(IterationCounter.advance())
else:
process = psutil.Process(os.getpid())
global_time_steps.append(ostime.time())
global_memory_step.append(process.memory_info().rss)
global_fds_step.append(len(psutil.Process().open_files()))
def perIterGC():
gc.collect()
if __name__ == '__main__':
field_chunksize = 1
do_chunking = False
auto_chunking = False
imageFileName = ""
parser = ArgumentParser(description="Example of particle advection around an idealised peninsula")
parser.add_argument("-f", "--fieldsize", dest="fieldsize", type=int, default=1, help="size of each field chunk")
parser.add_argument("-c", "--do-chunking", dest="do_chunking", action='store_true', default=False, help="enable/disable field chunking")
parser.add_argument("-a", "--auto-chunking", dest="auto_chunking", action='store_true', default=False, help="enable/disable auto-chunking")
parser.add_argument("-i", "--imageFileName", dest="imageFileName", type=str, default="mpiChunking_plot_MPI.png", help="image file name of the plot")
parser.add_argument("-b", "--backwards", dest="backwards", action='store_true', default=False, help="enable/disable running the simulation backwards")
parser.add_argument("-d", "--defer", dest="defer", action='store_false', default=True, help="enable/disable running with deferred load (default: True)")
parser.add_argument("-p", "--periodic", dest="periodic", action='store_true', default=False, help="enable/disable periodic wrapping (else: extrapolation)")
parser.add_argument("-r", "--repeatdt", dest="repeatdt", action='store_true', default=False, help="continuously add particles via repeatdt (default: False)")
args = parser.parse_args()
auto_chunking = args.auto_chunking
do_chunking = args.do_chunking
if auto_chunking:
do_chunking = True
elif do_chunking:
field_chunksize = args.fieldsize
imageFileName = args.imageFileName
deferLoadFlag = args.defer
periodicFlag = args.periodic
backwardSimulation = args.backwards
repeatdtFlag = args.repeatdt
odir = "/scratch/{}/experiments".format(os.environ['USER'])
func_time = []
mem_used_GB = []
open_fds = []
auto_field_size = 0
if MPI:
mpi_comm = MPI.COMM_WORLD
mpi_rank = mpi_comm.Get_rank()
mpi_size = mpi_comm.Get_size()
if mpi_rank == 0:
print("MPI - # workers: {}\n".format(mpi_size))
print("Dask global config - array.chunk-size: {}\n".format(da.config.get('array.chunk-size')))
else:
print("Dask global config - array.chunk-size: {}\n".format(da.config.get('array.chunk-size')))
if not do_chunking:
fieldset = set_cmems_fieldset(False, deferLoadFlag, periodicFlag)
elif auto_chunking:
fieldset = set_cmems_fieldset('auto', deferLoadFlag, periodicFlag)
else:
fieldset = set_cmems_fieldset(field_chunksize, deferLoadFlag, periodicFlag)
if MPI:
mpi_comm = MPI.COMM_WORLD
mpi_rank = mpi_comm.Get_rank()
process = psutil.Process(os.getpid())
if mpi_rank == 0:
global_t_0 = ostime.time()
global_m_0 = process.memory_info().rss
else:
process = psutil.Process(os.getpid())
global_t_0 = ostime.time()
global_m_0 = process.memory_info().rss
if with_ChunkInfoPrint:
if MPI:
mpi_comm = MPI.COMM_WORLD
mpi_comm.Barrier()
if mpi_comm.Get_rank() > 0:
pass
else:
print_field_info(fieldset)
else:
print_field_info()
simStart = None
for f in fieldset.get_fields():
if type(f) in [parcels.VectorField, parcels.NestedField, parcels.SummedField]:
continue
else:
if backwardSimulation:
simStart = f.grid.time_full[-1]
else:
simStart = f.grid.time_full[0]
break
if backwardSimulation:
# ==== backward simulation ==== #
if repeatdtFlag:
pset = ParticleSet(fieldset=fieldset, pclass=JITParticle, lon=np.random.rand(96, 1) * 1e-5, lat=np.random.rand(96, 1) * 1e-5, time=simStart, repeatdt=delta(hours=1))
else:
pset = ParticleSet(fieldset=fieldset, pclass=JITParticle, lon=np.random.rand(96, 1) * 1e-5, lat=np.random.rand(96, 1) * 1e-5, time=simStart)
else:
# ==== forward simulation ==== #
if repeatdtFlag:
pset = ParticleSet(fieldset=fieldset, pclass=JITParticle, lon=np.random.rand(96, 1) * 1e-5, lat=np.random.rand(96, 1) * 1e-5, repeatdt=delta(hours=1))
else:
pset = ParticleSet(fieldset=fieldset, pclass=JITParticle, lon=np.random.rand(96, 1) * 1e-5, lat=np.random.rand(96, 1) * 1e-5)
perflog = PerformanceLog()
postProcessFuncs = [perflog.advance, ]
if with_GC:
postProcessFuncs.append(perIterGC)
if backwardSimulation:
# ==== backward simulation ==== #
pset.execute(AdvectionRK4, runtime=delta(days=33), dt=delta(hours=-1), postIterationFunctions=postProcessFuncs)
else:
# ==== forward simulation ==== #
pset.execute(AdvectionRK4, runtime=delta(days=33), dt=delta(hours=1), postIterationFunctions=postProcessFuncs)
if auto_chunking:
if MPI:
mpi_comm = MPI.COMM_WORLD
mpi_rank = mpi_comm.Get_rank()
if mpi_rank == 0:
for f in fieldset.get_fields():
if type(f) in [parcels.VectorField, parcels.NestedField,
parcels.SummedField] or not f.grid.defer_load:
continue
chunk_info = f.grid.chunk_info
auto_field_size = chunk_info[chunk_info[0] + 1]
break
else:
for f in fieldset.get_fields():
if type(f) in [parcels.VectorField, parcels.NestedField, parcels.SummedField] or not f.grid.defer_load:
continue
chunk_info = f.grid.chunk_info
auto_field_size = chunk_info[chunk_info[0]+1]
break
if MPI:
mpi_comm = MPI.COMM_WORLD
mpi_comm.Barrier()
if mpi_comm.Get_rank() > 0:
pass
else:
plot(perflog.samples, perflog.times_steps, perflog.memory_steps, perflog.fds_steps, os.path.join(odir, imageFileName))
else:
plot(perflog.samples, perflog.times_steps, perflog.memory_steps, perflog.fds_steps, os.path.join(odir, imageFileName))