def simulate_bdc(lmrs, cache=None):
# BDC
print('BDC')
for lmr in lmrs:
optimize.set_wait_for(
lmr=lmr,
overhead=1.0,
decodingf=complexity.decoding0,
wait_for=1,
)
df = droplets.simulate(
partial(
delay.delay_mean_simulated,
overhead=1.0,
n=100,
order=delay.ROUND_ROBIN,
),
lmrs,
cache=cache,
)
# add decoding/straggling delay
for i, lmr in enumerate(lmrs):
if i >= len(df):
break
q, d = bdc.optimize_bdc(lmr)
tmp = df['delay'][i]
df.loc[i, 'delay'] += d
return df
def simulate_lt(lmrs):
for lmr in lmrs:
decodingf = partial(complexity.lt_complexity, reloverhead=1.3)
optimize.set_wait_for(
lmr=lmr,
overhead=1.3,
decodingf=decodingf,
)
return droplets.simulate(
partial(delay.delay_mean, overhead=1.3),
lmrs,
)
def simulate_ideal(lmrs):
for lmr in lmrs:
decodingf = complexity.decoding0
optimize.set_wait_for(
lmr=lmr,
overhead=1.0,
decodingf=decodingf,
wait_for=1,
)
return droplets.simulate(
partial(delay.delay_mean, overhead=1.0),
lmrs,
)
def approximate_r10(lmrs):
'''Approximation of R10 performance.'''
print('R10 approximation')
for lmr in lmrs:
decodingf = partial(complexity.r10_complexity, reloverhead=1.02)
optimize.set_wait_for(
lmr=lmr,
overhead=1.02,
decodingf=decodingf,
)
r10 = droplets.simulate(
partial(delay.delay_mean, overhead=1.020148),
lmrs,
)
return r10
def simulate_r10_opt(lmrs, cache=None):
'''R10 simulated, optimal droplet order.'''
for lmr in lmrs:
decodingf = partial(complexity.r10_complexity, reloverhead=1.02)
optimize.set_wait_for(
lmr=lmr,
overhead=1.02,
decodingf=decodingf,
# wait_for=int(round(lmr['nservers']/2))
)
print(lmr['wait_for'])
return droplets.simulate(
partial(
delay.delay_mean_empiric,
overhead=1.020148,
),
lmrs,
cache=cache,
)
def simulate_mds(lmrs):
print('MDS')
for lmr in lmrs:
decodingf = partial(
complexity.bdc_decoding_complexity,
code_length=lmr['nservers'],
partitions=1,
algorithm='fft',
)
optimize.set_wait_for(lmr=lmr,
overhead=1.0,
decodingf=decodingf,
wait_for=int(
round(lmr['code_rate'] * lmr['nservers'])))
df = droplets.simulate(
delay.delay_classical,
lmrs,
)
return df
def simulate_centralized_lt(lmrs):
print('LT cent.')
for lmr in lmrs:
decodingf = partial(
complexity.lt_complexity,
reloverhead=1.3,
)
optimize.set_wait_for(
lmr=lmr,
overhead=1.3,
decodingf=decodingf,
)
return droplets.simulate(
partial(
delay.delay_mean_centralized,
overhead=1.3,
),
lmrs,
)
def simulate_centralized_r10(lmrs):
print('R10 cent.')
for lmr in lmrs:
decodingf = partial(
complexity.r10_complexity,
reloverhead=1.02,
)
optimize.set_wait_for(
lmr=lmr,
overhead=1.02,
decodingf=decodingf,
)
df = droplets.simulate(
partial(
delay.delay_mean_centralized,
overhead=1.020148,
),
lmrs,
)
return df
def simulate_r10_rr(lmrs, cache=None, rerun=False):
'''R10 simulated, round-robin droplet order.'''
print('R10 simulated round-robin')
for lmr in lmrs:
decodingf = partial(complexity.r10_complexity, reloverhead=1.02)
optimize.set_wait_for(
lmr=lmr,
overhead=1.02,
decodingf=decodingf,
)
df = droplets.simulate(
partial(
delay.delay_mean_simulated,
overhead=1.020148,
n=100,
order=delay.ROUND_ROBIN,
),
lmrs,
cache=cache,
rerun=rerun,
)
return df
def test_optimize(self):
lmr = self.lmr1()
t1 = delay.delay_mean(lmr)
lmr = optimize.set_wait_for(
lmr=lmr,
overhead=1.02,
decodingf=complexity.testf,
)
t2 = delay.delay_mean(lmr)
# make sure optimization didn't increase the delay
self.assertLessEqual(t2, t1)
return
def lmr1(self):
lmr = typedefs.lmr_factory(
nservers=100,
nrows=100,
ncols=100,
nvectors=100,
ndroplets=200,
wait_for=80,
straggling_factor=1,
decodingf=lambda x: 0,
)
lmr = optimize.set_wait_for(
lmr=lmr,
overhead=1.02,
decodingf=complexity.testf,
wait_for=50,
)
return lmr