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 workload_plot():
cache_dir = './.simulate_cache/'
cache_prefix = 'workload_'
# get parameters to simulate
lmrs = get_parameters_workload()[:10]
# run simulations
uncoded = droplets.simulate(delay.delay_uncoded, lmrs)
mds = simulate_mds(lmrs)
mds['delay'] /= uncoded['delay']
bdc = simulate_bdc(
lmrs,
cache=path.join(cache_dir, cache_prefix + 'bdc'),
)
bdc['delay'] /= uncoded['delay']
r10 = approximate_r10(lmrs)
r10['delay'] /= uncoded['delay']
r10_rr = simulate_r10_rr(
lmrs,
cache=path.join(cache_dir, cache_prefix + 'r10_rr'),
)
r10_rr['delay'] /= uncoded['delay']
r10_opt = simulate_r10_opt(
lmrs,
# cache=path.join(cache_dir, cache_prefix+'r10_rr'),
)
r10_opt['delay'] /= uncoded['delay']
r10_cent = simulate_centralized_r10(lmrs)
r10_cent['delay'] /= uncoded['delay']
lt = simulate_lt(lmrs)
lt['delay'] /= uncoded['delay']
ideal = simulate_ideal(lmrs)
ideal['delay'] /= uncoded['delay']
# create plot
plt.figure()
plt.plot(
r10['nservers'],
r10['delay'],
r10_plot_string,
markevery=0.2,
markerfacecolor='none',
markeredgewidth=1.0,
label='R10',
)
plt.plot(
r10_opt['nservers'],
r10_opt['delay'],
# '--',
markevery=0.2,
markerfacecolor='none',
markeredgewidth=1.0,
label='R10 sim. (opt)',
)
plt.plot(
r10_rr['nservers'],
r10_rr['delay'],
'--',
markevery=0.2,
markerfacecolor='none',
markeredgewidth=1.0,
label='R10 sim. (rr)',
)
plt.plot(
lt['nservers'],
lt['delay'],
lt_plot_string,
markevery=0.2,
markerfacecolor='none',
markeredgewidth=1.0,
label='LT',
)
plt.plot(
ideal['nservers'],
ideal['delay'],
ideal_plot_string,
markevery=0.2,
markerfacecolor='none',
markeredgewidth=1.0,
label='Ideal Rateless.',
)
plt.plot(
bdc['nservers'],
bdc['delay'],
bdc_plot_string,
markevery=0.2,
markerfacecolor='none',
markeredgewidth=1.0,
label='BDC [6]',
)
plt.plot(
r10_cent['nservers'],
r10_cent['delay'],
centralized_plot_string,
markevery=0.2,
markerfacecolor='none',
markeredgewidth=1.0,
label='Cent. R10',
)
plt.plot(
mds['nservers'],
mds['delay'],
mds_plot_string,
markevery=0.2,
markerfacecolor='none',
markeredgewidth=1.0,
label='MDS [4]',
)
plt.grid(linestyle='--')
plt.legend(framealpha=1,
labelspacing=0.1,
columnspacing=0.1,
ncol=1,
loc='best')
plt.ylabel('$D$')
plt.xlabel('$K$')
plt.show()
return
def straggling_plot():
cache_dir = './.simulate_cache/'
cache_prefix = 'workload_'
# get parameters to simulate
lmrs = get_parameters_straggling()[:5]
# run simulations
uncoded = droplets.simulate(delay.delay_uncoded, lmrs)
mds = simulate_mds(lmrs)
mds['delay'] /= uncoded['delay']
bdc = simulate_bdc(
lmrs,
cache=path.join(cache_dir, cache_prefix + 'bdc'),
)
bdc['delay'] /= uncoded['delay']
r10 = approximate_r10(lmrs)
r10['delay'] /= uncoded['delay']
r10_rr = simulate_r10(
lmrs,
cache=path.join(cache_dir, cache_prefix + 'r10_rr'),
)
r10_rr['delay'] /= uncoded['delay']
# create plot
plt.figure()
plt.plot(
r10['nservers'],
r10['delay'],
r10_plot_string,
markevery=0.2,
markerfacecolor='none',
markeredgewidth=1.0,
label='R10',
)
plt.plot(
r10_rr['nservers'],
r10_rr['delay'],
'--',
markevery=0.2,
markerfacecolor='none',
markeredgewidth=1.0,
label='R10 sim. (rr)',
)
plt.plot(
bdc['nservers'],
bdc['delay'],
bdc_plot_string,
markevery=0.2,
label='BDC [6]',
)
plt.plot(
mds['nservers'],
mds['delay'],
classical_plot_string,
markevery=0.2,
markerfacecolor='none',
markeredgewidth=1.0,
label='Classical [4]',
)
plt.grid(linestyle='--')
plt.legend(framealpha=1,
labelspacing=0.1,
columnspacing=0.1,
ncol=1,
loc='best')
plt.ylabel('$D$')
plt.xlabel('$K$')
plt.show()
return