def create_all_histograms():
create_datadir_if_not_exists(GRAPHPATH)
keysize = "6"
shardings = [False, True]
_number_shardings = len(shardings)
middleware_latencies = np.zeros(
(_number_shardings, 2,
3)) # the "two" is for the number of middlewares
client_latencies = np.zeros(
(_number_shardings, 2, 3, 3)
) # the "two" and "three" is for middlewares and clients respectively
# Iterate through keysizes
for jdx, sharding in enumerate(shardings):
print("Sharding: ", sharding)
print("Keysize is: ", keysize)
# ONCE FOR THE MIDDLEWARE
for repetition in range(3):
for middleware in [1, 2]:
# Get throughput and latency from the file
middleware_filename = get_pattern_exp5_1_middleware(
middleware=middleware,
keysize=keysize,
repetition=repetition,
sharding=sharding)
try:
df = parse_log(BASEPATH + middleware_filename)
tmp_latencies_mw = get_histogram_latency_log_dataframe_middleware(
df).T
# print("ATT")
# print(jdx, idx, middleware, repetition)
# print(type(jdx), type(idx), type(middleware), type(repetition))
middleware_latencies[
jdx, idx, :, middleware - 1,
repetition] = get_avg_25_50_75_90_99_percentiles(
tmp_latencies_mw)
except Exception as e:
# print("WRONG WITH THE FOLLOWING CONFIG: ", client_filename)
print("WRONG WITH THE FOLLOWING CONFIG: ",
middleware_filename)
print(e)
continue
# ONCE FOR THE CLIENT
for client_idx, client in enumerate(
['Client1', 'Client2', 'Client3']):
for middleware in [1, 2]:
# TODO: Do it for each individual middleware!
# Trying to open and read the file!
client_filename = get_pattern_exp5_1_client(
keysize=keysize,
middleware=middleware,
repetition=repetition,
client=client,
sharding=sharding)
try:
_, tmp_latencies_client_get = read_client_histogram_as_dataframe(
filepath=BASEPATH + client_filename)
# print("At")
# print(jdx, idx, middleware, client_idx, repetition)
# print(type(jdx), type(idx), type(middleware), type(client_idx), type(repetition))
set_df, get_df = read_client_histogram_as_dataframe(
tmp_latencies_client_get)
# print("Indecies: ", jdx, idx, :, middleware - 1, client_idx, repetition)
if np.isnan(out).any():
print("Out is: ", out)
client_latencies[
jdx, idx, :, middleware - 1, client_idx,
repetition] = client_latencies[jdx, idx, :,
middleware - 1,
client_idx, 0]
else:
client_latencies[jdx, idx, :, middleware - 1,
client_idx, repetition] = out
except Exception as e:
# print("WRONG WITH THE FOLLOWING CONFIG: ", client_filename)
print("WRONG WITH THE FOLLOWING CONFIG: ",
middleware_filename)
print(e)
assert False
continue
# Normalizing the graphs to the appropriate shapes by taking means (and calculating stddevs)
mean_client_latencies = np.mean(client_latencies,
axis=1,
keepdims=True)
mean_client_latencies = np.mean(mean_client_latencies,
axis=2,
keepdims=True)
mean_middleware_latencies = np.mean(middleware_latencies,
axis=1,
keepdims=True)
mean_client_latencies = mean_client_latencies.squeeze()
mean_middleware_latencies = mean_middleware_latencies.squeeze()
print("Squeezed")
print(mean_client_latencies.shape)
print(mean_middleware_latencies.shape)
client_means = np.mean(mean_client_latencies[jdx, :, :, :], axis=2)
client_stddevs = np.std(mean_client_latencies[jdx, :, :, :], axis=2)
mw_means = np.mean(mean_middleware_latencies[jdx, :, :, :], axis=2)
mw_stddev = np.std(mean_middleware_latencies[jdx, :, :, :], axis=2)
print(mw_means)
print(client_means)
print("Means and Stddev")
print(client_means.shape)
print(client_stddevs.shape)
print(mw_means.shape)
print(mw_stddev.shape)
create_multiple_histogram_plot(
keys=multikeys,
means=client_means,
stddevs=client_stddevs,
filepath=GRAPHPATH +
"exp5_1_client_percentile_plots_sharded_{}".format(sharding))
create_multiple_histogram_plot(
keys=multikeys,
means=mw_means,
stddevs=mw_stddev,
filepath=GRAPHPATH +
"exp5_1_mw_percentile_plots_sharded_{}".format(sharding))
def iterate_through_experiments_exp3_2():
"""
Iterates through all the parameter combinations
:return:
"""
create_datadir_if_not_exists(GRAPHPATH)
middleware_threads = [(2 ** x) for x in range(3, 7)]
virtual_clients = [(2 ** x) for x in range(0, 6)]
total_virtual_clients = len([(2 ** x) for x in range(0, 6)])
total_middleware_threads = len([(2 ** x) for x in range(3, 7)])
for write in [0, 1]:
max_throughput = 0
max_throughput_client = 0
mw_throughput_means = np.zeros((total_virtual_clients, total_middleware_threads))
mw_throughput_stddev = np.zeros((total_virtual_clients, total_middleware_threads))
mw_latency_means = np.zeros((total_virtual_clients, total_middleware_threads))
mw_latency_stddev = np.zeros((total_virtual_clients, total_middleware_threads))
client_throughput_means = np.zeros((total_virtual_clients, total_middleware_threads))
client_latency_means = np.zeros((total_virtual_clients, total_middleware_threads))
client_throughput_stddev = np.zeros((total_virtual_clients, total_middleware_threads))
client_latency_stddev = np.zeros((total_virtual_clients, total_middleware_threads))
for _vc in range(0, 6):
vc = (2 ** _vc)
for _mt in range(3 - 3, 7 - 3):
mt = (2 ** (_mt + 3))
client_all_throughputs = [] # From here on, we average over all values
client_all_latencies = []
mw_all_throughputs = []
mw_all_latencies = []
queuetimes = []
for repetition in range(3):
for middleware in [1, 2]:
# Get throughput and latency from the file
middleware_filename = get_pattern_exp3_2_middleware(
virtualclientthreads=vc,
middlewarethreads=mt,
repetition=repetition,
write=write,
middleware=middleware
)
try:
df = parse_log(BASEPATH + middleware_filename)
mw_throughput, mw_latency = get_latency_log_dataframe(df)
queuetimes.append(df['differenceTimeEnqueuedAndDequeued'].astype(float).mean())
mw_throughput *= mt * 1000 # because we have two middlewares # * (2./3.) # because the load of the clients is distributed on two middlewares
print("Throughput is: ", mw_throughput)
mw_all_throughputs.append(mw_throughput)
mw_all_latencies.append(mw_latency)
except Exception as e:
print("WRONG WITH THE FOLLOWING CONFIG: ", middleware_filename)
print(e)
continue
for client in ['Client1', 'Client2', 'Client3']:
for middleware in [1, 2]:
client_filename = get_pattern_exp3_2_client(
virtual_client_threads=vc,
middlewareworkerthreads=mt,
repetition=repetition,
client=client,
write=write,
middleware=middleware
)
# Trying to open and read the file!
try:
client_throughput, client_latency = get_throughput_latency_default_memtier(filepath=BASEPATH + client_filename)
client_all_throughputs.append(client_throughput)
client_all_latencies.append(client_latency)
except Exception as e:
print("WRONG WITH THE FOLLOWING CONFIG: ", middleware_filename)
print(e)
continue
print("VC and MT ", (mt, vc))
print("NEW VC and MT ", (_mt, _vc))
mw_mean_latency = np.mean(mw_all_latencies)
mw_mean_throughput = np.sum(mw_all_throughputs) / 3. # because we have 3 repeats, and two middlewares (throughput per middleware is calculated as the total throughput)
mw_stddev_latency = np.std(mw_all_latencies)
mw_stddev_throughput = np.std(mw_all_throughputs)
if max_throughput < mw_mean_throughput:
max_throughput = mw_mean_throughput
print("TTT", write, mw_mean_throughput, mw_mean_latency, np.mean(queuetimes))
# Append to list
mw_throughput_means[_vc, _mt] = mw_mean_throughput
mw_latency_means[_vc, _mt] = mw_mean_latency
mw_throughput_stddev[_vc, _mt] = mw_stddev_throughput
mw_latency_stddev[_vc, _mt] = mw_stddev_latency
client_mean_latency = np.mean(client_all_latencies)
client_mean_throughput = np.sum(client_all_throughputs) / 3.
client_stddev_latency = np.std(client_all_latencies)
client_stddev_throughput = np.std(client_all_throughputs)
if max_throughput_client < client_mean_throughput:
max_throughput_client = client_mean_throughput
print("TTT CLIENT", write, client_mean_throughput, client_mean_latency)
# TTT 0 2978.673972968874 14.885140067475584 588816.9135975742
# TTT CLIENT 0 3010.066666666667 15.961666666666666
# Append to list
client_throughput_means[_vc, _mt] = client_mean_throughput
client_latency_means[_vc, _mt] = client_mean_latency
client_throughput_stddev[_vc, _mt] = client_stddev_throughput
client_latency_stddev[_vc, _mt] = client_stddev_latency
print(client_throughput_means[:,_mt].flatten())
print(total_virtual_clients)
# Assertions
len_mt = client_throughput_means[:, _mt].flatten().shape
len_vc = client_throughput_means[:, _mt].flatten().shape
len1 = virtual_clients
print("All lengths are : ")
print(len_mt)
print(len_vc)
print(len1)
# THROUGHPUTS
print("###write:client ", write, client_throughput_means)
print("###write:middleware ", write, mw_throughput_means)
###write:client 0
# [[ 1920.88 1918.63333333 1879.97666667 2040.63666667]
# [2948.59 2948.03 2953.48666667 4106.08333333]
# [2984.82333333 2906.10666667 2994.25 7329.56]
# [2968.63666667 2931.46333333 4117.91 9388.87666667]
# [2948.98 2924.93 10709.36 10652.36333333]
# [2967.09333333 2988.49 11116.91333333 10953.04]]
###write:middleware 0
# [[ 1823.82088122 1831.66427863 1788.32232092 1963.94693059]
# [2894.0011844 2893.19407003 2889.4399206 3919.63102998]
# [2904.84043756 2916.86120553 2914.29668445 6350.16067147]
# [2919.44080752 2923.44783548 2898.39238625 8313.77410761]
# [2916.26533345 2912.31134097 9796.16787827 9657.16189206]
# [2923.77401065 2917.48683475 10330.97247506 9984.91367675]]
###write:client 1
# [[1371.69 1433.82666667 1428.60333333 1474.68333333]
# [2243.24333333 2678.12 2598.63 2644.23]
# [3439.77333333 4868.33 4909.14 3774.43333333]
# [5648.88333333 5682.95333333 6149.03 6716.31666667]
# [5882.05 7103.95333333 6660.81333333 8108.17]
# [5883.05333333 6663.61333333 6983.25666667 4487.62]]
###write:middleware 1 [[1401.75925962 1472.40508414 1479.24123472 1517.59234247]
# [2784.22311061 2977.37234977 2626.56406027 2976.11177963]
# [5079.75686936 5304.33947114 4915.28801361 4549.81135794]
# [6597.81205289 6870.64256402 6942.7881017 6715.72352865]
# [6539.19650736 7693.89813362 8053.29642461 8131.41925129]
# [6670.31737933 7462.38036606 7429.34669134 7160.16151776]]
# LATENCIES
print("###write:client ", write, client_latency_means)
print("###write:middleware ", write, mw_latency_means)
###write:client 0
# [[ 1.57666667 1.58 1.61333333 1.49777778]
# [2.05555556 2.06222222 2.05777778 1.48888889]
# [4.05666667 4.17 4.03777778 1.65222222]
# [8.10111111 8.21222222 6.61666667 2.57333333]
# [16.22222222 16.40888889 4.49777778 4.53666667]
# [32.37111111 32.09222222 8.65888889 8.78444444]]
###write:middleware 0 [[ 0.75373062 0.71682754 0.79261609 0.68196475]
# [1.14245532 1.15601874 1.12761835 0.6962914]
# [3.00268178 3.06263071 3.08864053 0.82161966]
# [7.07679403 6.89052932 6.98758393 1.18371943]
# [15.26808786 15.11082288 1.48361568 1.50505344]
# [31.39592934 31.16684201 1.71758052 1.96693429]]
###write:client 1 [[ 2.18222222 2.08888889 2.09666667 2.03111111]
# [1.83222222 1.97444444 1.83111111 2.00444444]
# [2.02111111 2.17222222 1.92888889 1.63666667]
# [3.78 2.89777778 2.76444444 2.87222222]
# [7.29444444 5.97222222 5.64333333 3.69111111]
# [12.14444444 10.79 10.41555556 14.27666667]]
###write:middleware 1
# [[ 1.14301648 1.03218857 1.02287563 0.98682917]
# [1.02473659 1.01945742 1.08312422 1.00336116]
# [1.09529057 1.12426547 1.15199756 1.10026307]
# [2.38971881 1.71178708 1.66788647 1.66212348]
# [5.65331611 3.1920922 2.80038391 2.75255764]
# [11.12888022 4.42134564 2.9218675 2.73590902]]
render_lineargraph_multiple_errorbars(
labels=virtual_clients,
mean_array=client_throughput_means.T,
stddev_array=client_throughput_stddev.T,
filepath=GRAPHPATH + "exp3_2__throughput_client_write_{}".format(write),
is_latency=False
)
render_lineargraph_multiple_errorbars(
labels=virtual_clients,
mean_array=client_latency_means.T,
stddev_array=client_latency_stddev.T,
filepath=GRAPHPATH + "exp3_2__latency_client_write_{}".format(write),
is_latency=True
)
# Plot the middleware values
render_lineargraph_multiple_errorbars(
labels=virtual_clients,
mean_array=mw_throughput_means.T,
stddev_array=mw_throughput_stddev.T,
filepath=GRAPHPATH + "exp3_2__throughput_middleware_write_{}".format(write),
is_latency=False
)
render_lineargraph_multiple_errorbars(
labels=virtual_clients,
mean_array=mw_latency_means.T,
stddev_array=mw_latency_stddev.T,
filepath=GRAPHPATH + "exp3_2__latency_middleware_write_{}".format(write),
is_latency=True
)
def iterate_through_experiments_exp3_1():
"""
Iterates through all the parameter combinations
:return:
"""
create_datadir_if_not_exists(GRAPHPATH)
middleware_threads = [(2**x) for x in range(3, 7)]
virtual_clients = [(2**x) for x in range(0, 6)]
total_virtual_clients = len([(2**x) for x in range(0, 6)])
total_middleware_threads = len([(2**x) for x in range(3, 7)])
for write in [0, 1]:
max_throughput = 0
max_throughput_client = 0
mw_throughput_means = np.zeros(
(total_virtual_clients, total_middleware_threads))
mw_throughput_stddev = np.zeros(
(total_virtual_clients, total_middleware_threads))
mw_latency_means = np.zeros(
(total_virtual_clients, total_middleware_threads))
mw_latency_stddev = np.zeros(
(total_virtual_clients, total_middleware_threads))
client_throughput_means = np.zeros(
(total_virtual_clients, total_middleware_threads))
client_latency_means = np.zeros(
(total_virtual_clients, total_middleware_threads))
client_throughput_stddev = np.zeros(
(total_virtual_clients, total_middleware_threads))
client_latency_stddev = np.zeros(
(total_virtual_clients, total_middleware_threads))
for _vc in range(0, 6):
vc = (2**_vc)
for _mt in range(3 - 3, 7 - 3):
mt = (2**(_mt + 3))
client_all_throughputs = [
] # From here on, we average over all values
client_all_latencies = []
mw_all_throughputs = []
mw_all_latencies = []
queuetimes = []
for repetition in range(3):
# Get throughput and latency from the file
middleware_filename = get_pattern_exp3_1_middleware(
virtualclientthreads=vc,
middlewarethreads=mt,
repetition=repetition,
write=write,
)
try:
df = parse_log(BASEPATH + middleware_filename)
mw_throughput, mw_latency = get_latency_log_dataframe(
df)
mw_throughput *= mt * 1000 # bcs this gives us overall throughput per millisecond, and not per thread anymore
queuetimes.append(
df['differenceTimeEnqueuedAndDequeued'].astype(
float).mean())
mw_all_throughputs.append(mw_throughput)
mw_all_latencies.append(mw_latency)
except Exception as e:
print("WRONG WITH THE FOLLOWING CONFIG: ",
middleware_filename)
print(e)
continue
for client in ['Client1', 'Client2', 'Client3']:
client_filename = get_pattern_exp3_1_client(
virtual_client_threads=vc,
middlewareworkerthreads=mt,
repetition=repetition,
client=client,
write=write)
# Trying to open and read the file!
try:
client_throughput, client_latency = get_throughput_latency_default_memtier(
filepath=BASEPATH + client_filename)
client_all_throughputs.append(client_throughput)
client_all_latencies.append(client_latency)
print("Client: ", client_throughput,
client_latency)
except Exception as e:
print("WRONG WITH THE FOLLOWING CONFIG: ",
middleware_filename)
print(e)
continue
print("VC and MT ", (mt, vc))
print("NEW VC and MT ", (_mt, _vc))
mw_mean_latency = np.mean(mw_all_latencies)
mw_mean_throughput = np.sum(mw_all_throughputs) / 3.
mw_stddev_latency = np.std(mw_all_latencies)
mw_stddev_throughput = np.std(mw_all_throughputs)
if max_throughput < mw_mean_throughput:
max_throughput = mw_mean_throughput
print("TTT", write, mw_mean_throughput, mw_mean_latency,
np.mean(queuetimes))
# Append to list
mw_throughput_means[_vc, _mt] = mw_mean_throughput
mw_latency_means[_vc, _mt] = mw_mean_latency
mw_throughput_stddev[_vc, _mt] = mw_stddev_throughput
mw_latency_stddev[_vc, _mt] = mw_stddev_latency
client_mean_latency = np.mean(client_all_latencies)
client_mean_throughput = np.sum(client_all_throughputs) / 3.
client_stddev_latency = np.std(client_all_latencies)
client_stddev_throughput = np.std(client_all_throughputs)
if max_throughput_client < client_mean_throughput:
max_throughput_client = client_mean_throughput
print("TTT CLIENT", write, client_mean_throughput,
client_mean_latency)
# Maximal values at: (printouts!)
# Append to list
client_throughput_means[_vc, _mt] = client_mean_throughput
client_latency_means[_vc, _mt] = client_mean_latency
client_throughput_stddev[_vc, _mt] = client_stddev_throughput
client_latency_stddev[_vc, _mt] = client_stddev_latency
print(client_throughput_means[:, _mt].flatten())
print(total_virtual_clients)
# Assertions
len_mt = client_throughput_means[:, _mt].flatten().shape
len_vc = client_throughput_means[:, _mt].flatten().shape
len1 = virtual_clients
print("All lengths are : ")
print(len_mt)
print(len_vc)
print(len1)
# THROUGHPUTS
print("###write:client ", write, client_throughput_means)
print("###write:middleware ", write, mw_throughput_means)
# LATENCIES
print("###write:client ", write, client_latency_means)
print("###write:middleware ", write, mw_latency_means)
###write:client 0 [[2829.99666667 2828.78 2841.12333333 2832.09666667]
# [2914.32 2916.49 2902.36666667 2963.79 ]
# [2978.07 2982.04 2926.90666667 2961.14 ]
# [2930.64 2996.5 2980.55333333 2971.27666667]
# [2920.55666667 2939.13 2975.22333333 2949.03333333]
# [2966.08666667 2969.74666667 2933.78 2944.79 ]]
# ###write:middleware 0 [[2888.56768177 2872.94686583 2876.40834308 2863.0490837 ]
# [2952.04131319 2937.70233791 2918.01762357 2947.80422581]
# [2989.20268883 2974.35714111 2912.9803837 2934.38863854]
# [2933.52705812 2979.39794505 2961.4934029 2940.88512509]
# [2915.57629929 2930.44136708 2952.3195043 2941.34950968]
# [2983.28945045 2974.97495859 2925.1779705 2935.00349311]]
# ###write:client 0 [[ 2.11333333 2.11222222 2.10555556 2.11111111]
# [ 4.12222222 4.11555556 4.14 4.04888889]
# [ 8.06 8.05444444 8.19444444 8.10111111]
# [16.44666667 16.03333333 16.06111111 16.17111111]
# [32.96 32.63777778 32.23333333 32.51333333]
# [64.82444444 64.60222222 65.45 65.20888889]]
# ###write:middleware 0 [[ 1.3167424 1.18404118 1.19176319 1.17406692]
# [ 3.05772075 3.06435006 3.08202826 3.02690768]
# [ 7.03875969 6.86615187 6.97812098 6.82174604]
# [15.12661499 14.78340234 14.6029601 14.49839125]
# [31.04134367 30.77606178 30.94594595 30.51061114]
# [64.37984496 63.11079911 63.46138996 62.44980695]]
###write:client 1 [[3271.34 3220.32666667 3215.53333333 3321.11666667]
# [5386.31666667 5549.33333333 5136.30666667 5292.51333333]
# [6729.64333333 6900.06333333 6690.59666667 6214.48333333]
# [7061.56 7363.10333333 7541.20666667 7139.32333333]
# [7295.09 7490.80666667 7629.65333333 6914.55 ]
# [7228.72 6711.01666667 7753.10333333 7454.68 ]]
# ###write:middleware 1 [[3287.41072931 3284.10276218 3236.10885263 3348.17928571]
# [5425.01178366 5583.56806439 5174.72885707 5332.7263551 ]
# [6667.75480504 6881.43820305 6818.78960064 7011.40707111]
# [6968.99163433 7292.81531008 7476.06856816 7092.02329642]
# [7199.49091298 7430.69321895 7658.79654156 7739.66985074]
# [7369.68367374 7615.74155019 7635.26281836 7333.33824439]]
# ###write:client 1 [[ 1.83333333 1.86333333 1.86444444 1.80555556]
# [ 2.24777778 2.16444444 2.34555556 2.28 ]
# [ 3.58 3.48222222 3.48888889 3.41111111]
# [ 6.81444444 6.53555556 6.38222222 6.77 ]
# [13.19111111 12.80444444 12.57444444 12.30222222]
# [26.64888889 25.42666667 24.74444444 25.82111111]]
# ###write:middleware 1 [[ 0.95865633 0.99356499 0.98584299 0.94144144]
# [ 1.05426357 1.07722008 1.14740262 1.10841916]
# [ 1.73901809 1.4002574 1.41508969 1.48529033]
# [ 3.26873385 1.85971686 1.72651223 1.84685326]
# [ 5.56330749 2.57014157 2.15837807 2.01412311]
# [15.27655236 3.07501152 2.19581187 2.69169728]]
render_lineargraph_multiple_errorbars(
labels=virtual_clients,
mean_array=client_throughput_means.T,
stddev_array=client_throughput_stddev.T,
filepath=GRAPHPATH +
"exp3_1__throughput_client_write_{}".format(write),
is_latency=False)
render_lineargraph_multiple_errorbars(
labels=virtual_clients,
mean_array=client_latency_means.T,
stddev_array=client_latency_stddev.T,
filepath=GRAPHPATH +
"exp3_1__latency_client_write_{}".format(write),
is_latency=True)
# Plot the middleware values
render_lineargraph_multiple_errorbars(
labels=virtual_clients,
mean_array=mw_throughput_means.T,
stddev_array=mw_throughput_stddev.T,
filepath=GRAPHPATH +
"exp3_1__throughput_middleware_write_{}".format(write),
is_latency=False)
render_lineargraph_multiple_errorbars(
labels=virtual_clients,
mean_array=mw_latency_means.T,
stddev_array=mw_latency_stddev.T,
filepath=GRAPHPATH +
"exp3_1__latency_middleware_write_{}".format(write),
is_latency=True)
def iterate_through_experiments_exp3_2():
"""
Iterates through all the parameter combinations
:return:
"""
create_datadir_if_not_exists(GRAPHPATH)
middleware_threads = [(2**x) for x in range(3, 7)]
virtual_clients = [(2**x) for x in range(0, 6)]
total_virtual_clients = len([(2**x) for x in range(0, 6)])
total_middleware_threads = len([(2**x) for x in range(3, 7)])
for write in [1]:
mw_queueusize_means = np.zeros(
(total_virtual_clients, total_middleware_threads))
mw_queueusize_stddev = np.zeros(
(total_virtual_clients, total_middleware_threads))
for _vc in range(0, 6):
vc = (2**_vc)
for _mt in range(3 - 3, 7 - 3):
mt = (2**(_mt + 3))
all_queuesizes = []
for repetition in range(3):
for middleware in [1, 2]:
# Get throughput and latency from the file
middleware_filename = get_pattern_exp4_1_middleware(
virtualclientthreads=vc,
middlewarethreads=mt,
repetition=repetition,
write=write,
middleware=middleware)
try:
df = parse_log(BASEPATH + middleware_filename)
avg_queuesize = get_average_queuesize(df)
all_queuesizes.append(avg_queuesize)
except Exception as e:
print("WRONG WITH THE FOLLOWING CONFIG: ",
middleware_filename)
print(e)
continue
print("VC and MT ", (mt, vc))
print("NEW VC and MT ", (_mt, _vc))
mw_mean_queuesize = np.mean(all_queuesizes)
mw_stddev_queuesize = np.std(all_queuesizes)
# Append to list
mw_queueusize_means[_vc, _mt] = mw_mean_queuesize
mw_queueusize_stddev[_vc, _mt] = mw_stddev_queuesize
# Plot the middleware values
render_lineargraph_multiple_errorbars(
labels=virtual_clients,
mean_array=mw_queueusize_means.T,
stddev_array=mw_queueusize_stddev.T,
filepath=GRAPHPATH +
"exp4_1__queuesize_middleware_write_{}".format(write),
is_latency=False,
is_queue_size=True)
def create_queue_barplots():
create_datadir_if_not_exists(GRAPHPATH)
middleware_threads = [(2**x) for x in range(3, 7)]
virtual_clients = [(2**x) for x in range(0, 6)]
total_virtual_clients = len([(2**x) for x in range(0, 6)])
total_middleware_threads = len([(2**x) for x in range(3, 7)])
# The following are the different values we're going to plot:
#
time_labels = [
'Time to Enqueue', 'Time in Queue', 'Time Queue to Server',
'Time at Server', 'Time Server to Client'
]
print("Time labels have shape: ", len(time_labels))
for write in [0, 1]:
for _vc in range(0, 6):
vc = (2**_vc)
middleware_latencies = np.zeros(
(total_middleware_threads, len(time_labels), 1,
3)) # 3 repetitions, 2 middlewares
for _mt, mt in enumerate(middleware_threads):
# ONCE FOR THE MIDDLEWARE
for repetition in range(3):
for middleware in [1]:
# Get throughput and latency from the file
middleware_filename = get_pattern_exp3_1_middleware(
virtualclientthreads=vc,
middlewarethreads=mt,
repetition=repetition,
write=write,
)
try:
df = parse_log(BASEPATH + middleware_filename)
middleware_latencies[
_mt, :, middleware - 1,
repetition] = get_average_queue_components(df)
# print("item is: ", middleware_latencies[_mt, :, middleware - 1, repetition])
# print(middleware_latencies)
except Exception as e:
# print("WRONG WITH THE FOLLOWING CONFIG: ", client_filename)
print("WRONG WITH THE FOLLOWING CONFIG: ",
middleware_filename)
print(e)
continue
# Normalizing the graphs to the appropriate shapes by taking means (and calculating stddevs)
mean_middleware_latencies = np.mean(middleware_latencies,
axis=2,
keepdims=True)
stddev_middleware_latencies = np.std(mean_middleware_latencies,
axis=3,
keepdims=True)
mean_middleware_latencies = np.mean(mean_middleware_latencies,
axis=3,
keepdims=True)
mw_means = mean_middleware_latencies.squeeze().T
mw_stddev = stddev_middleware_latencies.squeeze().T
print("Means and Stddev")
print(mw_means.shape)
print(mw_stddev.shape)
print(len(time_labels))
print(time_labels)
print(mw_means)
create_multiple_histogram_plot(
keys=middleware_threads,
means=mw_means.T,
stddevs=mw_stddev.T,
filepath=GRAPHPATH +
"exp3_1_mw_percentile_plots_writes_{}__vc_{}".format(
write, vc),
is_queue=True)
def iterate_through_experiments_exp4_1():
"""
Iterates through all the parameter combinations
:return:
"""
create_datadir_if_not_exists(GRAPHPATH)
write = 1 # This experiment consists of writes-only!
middleware_threads = [(2**x) for x in range(3, 7)]
virtual_clients = [(2**x) for x in range(0, 6)]
total_virtual_clients = len([(2**x) for x in range(0, 6)])
total_middleware_threads = len([(2**x) for x in range(3, 7)])
mw_throughput_means = np.zeros(
(total_virtual_clients, total_middleware_threads))
mw_throughput_stddev = np.zeros(
(total_virtual_clients, total_middleware_threads))
mw_latency_means = np.zeros(
(total_virtual_clients, total_middleware_threads))
mw_latency_stddev = np.zeros(
(total_virtual_clients, total_middleware_threads))
client_throughput_means = np.zeros(
(total_virtual_clients, total_middleware_threads))
client_latency_means = np.zeros(
(total_virtual_clients, total_middleware_threads))
client_throughput_stddev = np.zeros(
(total_virtual_clients, total_middleware_threads))
client_latency_stddev = np.zeros(
(total_virtual_clients, total_middleware_threads))
queuetime_means = np.zeros(
(total_virtual_clients, total_middleware_threads))
queuelength_means = np.zeros(
(total_virtual_clients, total_middleware_threads))
waittime_means = np.zeros(
(total_virtual_clients, total_middleware_threads))
for _mt in range(3 - 3, 7 - 3):
mt = (2**(_mt + 3))
for _vc in range(0, 6):
vc = (2**_vc)
client_all_throughputs = [
] # From here on, we average over all values
client_all_latencies = []
mw_all_throughputs = []
mw_all_latencies = []
queuetimes = []
queuelengths = []
waittimes = []
for repetition in range(3):
for middleware in [1, 2]:
# Get throughput and latency from the file
middleware_filename = get_pattern_exp4_1_middleware(
virtualclientthreads=vc,
middlewarethreads=mt,
repetition=repetition,
write=write,
middleware=middleware)
try:
df = parse_log(BASEPATH + middleware_filename)
# Append queuetimes, queuelength and waittime
queuetimes.append(
df['differenceTimeEnqueuedAndDequeued'].astype(
float).mean())
queuelengths.append(
df['queueSize'].astype(float).mean())
waittimes.append(df[
'differenceTimeSentToServerAndReceivedResponseFromServer']
.astype(float).mean())
mw_throughput, mw_latency = get_latency_log_dataframe(
df)
mw_throughput *= mt * 1000 # bcs this gives us throughput per millisecond
# print("THROUGHPUT AND LATENCY ARE: ")
# print(mw_throughput)
# print(mw_latency)
mw_all_throughputs.append(mw_throughput)
mw_all_latencies.append(mw_latency)
except Exception as e:
# print("WRONG WITH THE FOLLOWING CONFIG: ", client_filename)
print("WRONG WITH THE FOLLOWING CONFIG: ",
middleware_filename)
print(e)
continue
for client in ['Client1', 'Client2', 'Client3']:
for middleware in [1, 2]:
# Trying to open and read the file!
client_filename = get_pattern_exp4_1_client(
virtual_client_threads=vc,
middleware=middleware,
middlewareworkerthreads=mt,
repetition=repetition,
client=client,
write=write)
try:
client_throughput, client_latency = get_throughput_latency_default_memtier(
filepath=BASEPATH + client_filename)
client_all_throughputs.append(client_throughput)
client_all_latencies.append(client_latency)
except Exception as e:
# print("WRONG WITH THE FOLLOWING CONFIG: ", client_filename)
print("WRONG WITH THE FOLLOWING CONFIG: ",
middleware_filename)
print(e)
continue
# Queuetimes etc.
queuetime_means[_vc, _mt] = np.mean(queuetimes)
queuelength_means[_vc, _mt] = np.mean(queuelengths)
waittime_means[_vc, _mt] = np.mean(waittimes)
mw_mean_latency = np.mean(mw_all_latencies)
mw_mean_throughput = np.sum(mw_all_throughputs) / 3.
mw_stddev_latency = np.std(mw_all_latencies)
mw_stddev_throughput = np.std(mw_all_throughputs)
# Append to list
mw_throughput_means[_vc, _mt] = mw_mean_throughput
mw_latency_means[_vc, _mt] = mw_mean_latency
mw_throughput_stddev[_vc, _mt] = mw_stddev_throughput
mw_latency_stddev[_vc, _mt] = mw_stddev_latency
client_mean_latency = np.mean(client_all_latencies)
client_mean_throughput = np.sum(client_all_throughputs) / 3.
client_stddev_latency = np.std(client_all_latencies)
client_stddev_throughput = np.std(client_all_throughputs)
# Append to list
client_throughput_means[_vc, _mt] = client_mean_throughput
client_latency_means[_vc, _mt] = client_mean_latency
client_throughput_stddev[_vc, _mt] = client_stddev_throughput
client_latency_stddev[_vc, _mt] = client_stddev_latency
# Assertions
len_mt = client_throughput_means[:, _mt].flatten().shape
len_vc = client_throughput_means[:, _mt].flatten().shape
len1 = virtual_clients
argmax_vc = np.argmax(mw_throughput_means, axis=0)
print("MW throughputs are: ", mw_throughput_means)
print("VC maximizing is: ", argmax_vc)
for idx, mt in enumerate([8, 16, 32, 64]):
max_vc = argmax_vc[idx]
print()
print()
print()
print("MT IS: ", mt)
# Section 7 data
# print("###write:client ", client_throughput_means[max_vc, idx])
# print("###write:middleware ", mw_throughput_means[max_vc, idx])
print("###write:queuesizes ", queuelength_means[max_vc, idx])
print("###write:latency ", mw_latency_means[max_vc, idx])
print("###write:waittime_means ",
waittime_means[max_vc, idx] / 1000000.)
###write:client 1
# [[ 1815.45333333 2321.54 2285.64 2127.40333333]
# [ 4608.49 4581.64 4576.45333333 4406.16333333]
# [ 6971.06666667 6543.16333333 7374.53333333 6144.32333333]
# [ 8698.22666667 8063.80666667 9086.58 9738.00333333]
# [ 8995.92 10068.55333333 11404.78 10261.27 ]
# [ 7064.09333333 10026.83333333 12586.12 12858.95333333]]
# ###write:middleware 1
# [[ 1947.229088 2317.76063679 2322.53675249 2207.33621436]
# [ 4591.30620045 4605.04654336 4582.13005631 4509.73165549]
# [ 8020.03037995 6967.31991722 7475.40874085 7804.08509258]
# [ 8645.69236829 10314.9812626 9302.41430409 9688.76860568]
# [ 8833.92859926 10153.21361682 12126.18091513 10299.9702301 ]
# [ 7465.56068924 10133.52159323 12503.89763076 13706.73554513]]
# Taken out for experiment 7!
# Plot the client values
render_lineargraph_multiple_errorbars(
labels=virtual_clients,
mean_array=client_throughput_means.T,
stddev_array=client_throughput_stddev.T,
filepath=GRAPHPATH +
"exp4_1__vc_{}__throughput_client_write_{}".format(mt, write),
is_latency=False)
render_lineargraph_multiple_errorbars(
labels=virtual_clients,
mean_array=client_latency_means.T,
stddev_array=client_latency_stddev.T,
filepath=GRAPHPATH +
"exp4_1__vc_{}__latency_client_write_{}".format(mt, write),
is_latency=True)
# print(np.max(mw_throughput_means, axis=0))
# print(np.max(mw_throughput_means, axis=1))
# Plot the middleware values
render_lineargraph_multiple_errorbars(
labels=virtual_clients,
mean_array=mw_throughput_means.T,
stddev_array=mw_throughput_stddev.T,
filepath=GRAPHPATH +
"exp4_1__vc_{}__throughput_middleware_write_{}".format(mt, write),
is_latency=False)
render_lineargraph_multiple_errorbars(
labels=virtual_clients,
mean_array=mw_latency_means.T,
stddev_array=mw_latency_stddev.T,
filepath=GRAPHPATH +
"exp4_1__vc_{}__latency_middleware_write_{}".format(mt, write),
is_latency=True)
def iterate_through_experiments_exp4_1():
"""
Iterates through all the parameter combinations
:return:
"""
create_datadir_if_not_exists(GRAPHPATH)
write = 1 # This experiment consists of writes-only!
multikeys = ["1", "3", "6", "9"]
shardings = [False, True]
_number_keys = len(multikeys)
_number_shardings = len(shardings)
mw_throughput_means = np.zeros((_number_keys, _number_shardings))
mw_throughput_stddev = np.zeros((_number_keys, _number_shardings))
mw_latency_means = np.zeros((_number_keys, _number_shardings))
mw_latency_stddev = np.zeros((_number_keys, _number_shardings))
client_throughput_means = np.zeros((_number_keys, _number_shardings))
client_latency_means = np.zeros((_number_keys, _number_shardings))
client_throughput_stddev = np.zeros((_number_keys, _number_shardings))
client_latency_stddev = np.zeros((_number_keys, _number_shardings))
# Iterate through keysizes
for jdx, sharding in enumerate(shardings):
print("Sharding: ", sharding)
for idx, keysize in enumerate(multikeys):
print("Keysize is: ", keysize)
client_all_throughputs = [
] # From here on, we average over all values
client_all_latencies = []
mw_all_throughputs = []
mw_all_latencies = []
for repetition in range(3):
for middleware in [1, 2]:
# Get throughput and latency from the file
middleware_filename = get_pattern_exp5_1_middleware(
middleware=middleware,
keysize=keysize,
repetition=repetition,
sharding=sharding)
try:
df = parse_log(BASEPATH + middleware_filename)
mw_throughput, mw_latency = get_latency_log_dataframe(
df)
mw_throughput *= 64 * 1000 # bcs this gives us throughput per millisecond
# print("THROUGHPUT AND LATENCY ARE: ")
# print(mw_throughput)
# print(mw_latency)
mw_all_throughputs.append(mw_throughput)
mw_all_latencies.append(mw_latency)
except Exception as e:
# print("WRONG WITH THE FOLLOWING CONFIG: ", client_filename)
print("WRONG WITH THE FOLLOWING CONFIG: ",
middleware_filename)
print(e)
continue
for client in ['Client1', 'Client2', 'Client3']:
for middleware in [1, 2]:
# TODO: Do it for each individual middleware!
# Trying to open and read the file!
client_filename = get_pattern_exp5_1_client(
keysize=keysize,
middleware=middleware,
repetition=repetition,
client=client,
sharding=sharding)
try:
client_throughput, client_latency = get_throughput_latency_default_memtier(
filepath=BASEPATH + client_filename)
client_all_throughputs.append(client_throughput)
client_all_latencies.append(client_latency)
except Exception as e:
# print("WRONG WITH THE FOLLOWING CONFIG: ", client_filename)
print("WRONG WITH THE FOLLOWING CONFIG: ",
middleware_filename)
print(e)
continue
mw_mean_latency = np.mean(mw_all_latencies)
mw_mean_throughput = np.sum(mw_all_throughputs) / 3.
mw_stddev_latency = np.std(mw_all_latencies)
mw_stddev_throughput = np.std(mw_all_throughputs)
# Append to list
mw_throughput_means[idx, jdx] = mw_mean_throughput
mw_latency_means[idx, jdx] = mw_mean_latency
mw_throughput_stddev[idx, jdx] = mw_stddev_throughput
mw_latency_stddev[idx, jdx] = mw_stddev_latency
client_mean_latency = np.mean(client_all_latencies)
client_mean_throughput = np.sum(client_all_throughputs) / 3.
client_stddev_latency = np.std(client_all_latencies)
client_stddev_throughput = np.std(client_all_throughputs)
# Append to list
client_throughput_means[idx, jdx] = client_mean_throughput
client_latency_means[idx, jdx] = client_mean_latency
client_throughput_stddev[idx, jdx] = client_stddev_throughput
client_latency_stddev[idx, jdx] = client_stddev_latency
# Middleware
render_lineargraph_errorbars(
labels=multikeys,
mean_array=mw_latency_means[:, jdx],
stddev_array=mw_latency_stddev[:, jdx],
filepath=GRAPHPATH +
"exp5_1__mw_latency_sharding_{}".format(sharding),
is_latency=True)
# Client
render_lineargraph_errorbars(
labels=multikeys,
mean_array=client_latency_means[:, jdx],
stddev_array=client_latency_stddev[:, jdx],
filepath=GRAPHPATH +
"exp5_1__client_latency_sharding_{}".format(sharding),
is_latency=True)
# Middleware
render_lineargraph_errorbars(
labels=multikeys,
mean_array=mw_throughput_means[:, jdx],
stddev_array=mw_throughput_stddev[:, jdx],
filepath=GRAPHPATH +
"exp5_1__mw_throughput_sharding_{}".format(sharding),
is_latency=False)
# Client
render_lineargraph_errorbars(
labels=multikeys,
mean_array=client_throughput_means[:, jdx],
stddev_array=client_throughput_stddev[:, jdx],
filepath=GRAPHPATH +
"exp5_1__client_throughput_sharding_{}".format(sharding),
is_latency=False)
def iterate_through_experiments_exp6_1():
"""
Iterates through all the parameter combinations
:return:
"""
write = 1 # This experiment consists of writes-only!
final_client_throughputs = []
final_mw_throughputs = []
final_client_latencies = []
final_mw_latencies = []
print("WRITES ARE:::: ", write)
for number_of_middlewares in [1, 2]:
for number_of_servers in [1, 3]:
for number_of_middleware_workerthreads in [8, 32]:
client_all_throughputs = [
] # From here on, we average over all values
client_all_latencies = []
for repetition in range(3):
mw_all_throughputs = []
mw_all_latencies = []
_array_middlewares = [
1, 2
] if number_of_middlewares == 2 else [1]
for middleware in _array_middlewares:
# Get throughput and latency from the file
middleware_filename = get_pattern_exp6_1_middleware(
middlewarethreads=
number_of_middleware_workerthreads,
number_of_middlewares=number_of_middlewares,
servers=number_of_servers,
repetition=repetition,
write=write,
middleware=middleware)
try:
df = parse_log(BASEPATH + middleware_filename)
mw_throughput, mw_latency = get_latency_log_dataframe(
df)
mw_throughput *= number_of_middleware_workerthreads * 1000 # bcs this gives us throughput per millisecond
# print("THROUGHPUT AND LATENCY ARE: ")
# print(mw_throughput)
# print(mw_latency)
mw_all_throughputs.append(mw_throughput)
mw_all_latencies.append(mw_latency)
except Exception as e:
# print("WRONG WITH THE FOLLOWING CONFIG: ", client_filename)
print("WRONG WITH THE FOLLOWING CONFIG: ",
middleware_filename)
print(e)
continue
configuration = (number_of_middlewares, number_of_servers,
number_of_middleware_workerthreads)
final_mw_throughputs.append(
(configuration, (repetition,
np.sum(mw_all_throughputs))))
final_mw_latencies.append(
(configuration, (repetition,
np.mean(mw_all_latencies))))
# Iterating through individual lists for easy copy pasta
# print("CLIENT THROUGHPUT")
# for x in final_client_throughputs:
# print(x)
print("MIDDLEWARE THROUGHPUT")
for idx, x in enumerate(final_mw_throughputs):
if idx % 3 == 0:
print()
print(x)
# print("CLIENT LATENCY")
# for x in final_client_latencies:
# print(x)
print("MIDDLEWARE LATENCY")
for idx, x in enumerate(final_mw_latencies):
if idx % 3 == 0:
print()
print(x)
def iterate_through_experiments_exp4_1():
"""
Iterates through all the parameter combinations
:return:
"""
create_datadir_if_not_exists(GRAPHPATH)
write = 1 # This experiment consists of writes-only!
middleware_threads = [(2**x) for x in range(3, 7)]
virtual_clients = [(2**x) for x in range(0, 6)]
total_virtual_clients = len([(2**x) for x in range(0, 6)])
total_middleware_threads = len([(2**x) for x in range(3, 7)])
mw_throughput_means = np.zeros(
(total_virtual_clients, total_middleware_threads))
mw_throughput_stddev = np.zeros(
(total_virtual_clients, total_middleware_threads))
mw_latency_means = np.zeros(
(total_virtual_clients, total_middleware_threads))
mw_latency_stddev = np.zeros(
(total_virtual_clients, total_middleware_threads))
client_throughput_means = np.zeros(
(total_virtual_clients, total_middleware_threads))
client_latency_means = np.zeros(
(total_virtual_clients, total_middleware_threads))
client_throughput_stddev = np.zeros(
(total_virtual_clients, total_middleware_threads))
client_latency_stddev = np.zeros(
(total_virtual_clients, total_middleware_threads))
for _mt in range(3 - 3, 7 - 3):
mt = (2**(_mt + 3))
queuetimes = []
queuesizes = []
waittimes = []
for _vc in range(0, 6):
vc = (2**_vc)
client_all_throughputs = [
] # From here on, we average over all values
client_all_latencies = []
mw_all_throughputs = []
mw_all_latencies = []
tmp_queuetimes = []
tmp_queuesizes = []
tmp_waittimes = []
for repetition in range(3):
for middleware in [1, 2]:
# Get throughput and latency from the file
middleware_filename = get_pattern_exp4_1_middleware(
virtualclientthreads=vc,
middlewarethreads=mt,
repetition=repetition,
write=write,
middleware=middleware)
try:
df = parse_log(BASEPATH + middleware_filename)
queuetime = df[
'differenceTimeEnqueuedAndDequeued'].astype(
float).mean() / 1000.
tmp_queuetimes.append(queuetime)
queuesize = df['queueSize'].astype(float).mean()
tmp_queuesizes.append(queuesize)
waittime = df[
'differenceTimeSentToServerAndReceivedResponseFromServer'].astype(
float).mean() / 1000.
tmp_waittimes.append(waittime)
mw_throughput, mw_latency = get_latency_log_dataframe(
df)
mw_throughput *= mt * 1000 # bcs this gives us throughput per millisecond
# print("THROUGHPUT AND LATENCY ARE: ")
# print(mw_throughput)
# print(mw_latency)
mw_all_throughputs.append(mw_throughput)
mw_all_latencies.append(mw_latency)
except Exception as e:
# print("WRONG WITH THE FOLLOWING CONFIG: ", client_filename)
print("WRONG WITH THE FOLLOWING CONFIG: ",
middleware_filename)
print(e)
continue
for client in ['Client1', 'Client2', 'Client3']:
for middleware in [1, 2]:
# Trying to open and read the file!
client_filename = get_pattern_exp4_1_client(
virtual_client_threads=vc,
middleware=middleware,
middlewareworkerthreads=mt,
repetition=repetition,
client=client,
write=write)
try:
client_throughput, client_latency = get_throughput_latency_default_memtier(
filepath=BASEPATH + client_filename)
client_all_throughputs.append(client_throughput)
client_all_latencies.append(client_latency)
except Exception as e:
# print("WRONG WITH THE FOLLOWING CONFIG: ", client_filename)
print("WRONG WITH THE FOLLOWING CONFIG: ",
middleware_filename)
print(e)
continue
queuetimes.append(np.mean(tmp_queuetimes))
queuesizes.append(np.mean(tmp_queuesizes))
waittimes.append(np.mean(tmp_waittimes))
# print("VC and MT ", (mt, vc))
# print("NEW VC and MT ", (_mt, _vc))
mw_mean_latency = np.mean(mw_all_latencies)
mw_mean_throughput = np.sum(mw_all_throughputs) / 3.
mw_stddev_latency = np.std(mw_all_latencies)
mw_stddev_throughput = np.std(mw_all_throughputs)
# Append to list
mw_throughput_means[_vc, _mt] = mw_mean_throughput
mw_latency_means[_vc, _mt] = mw_mean_latency
mw_throughput_stddev[_vc, _mt] = mw_stddev_throughput
mw_latency_stddev[_vc, _mt] = mw_stddev_latency
client_mean_latency = np.mean(client_all_latencies)
client_mean_throughput = np.sum(client_all_throughputs) / 3.
client_stddev_latency = np.std(client_all_latencies)
client_stddev_throughput = np.std(client_all_throughputs)
# Append to list
client_throughput_means[_vc, _mt] = client_mean_throughput
client_latency_means[_vc, _mt] = client_mean_latency
client_throughput_stddev[_vc, _mt] = client_stddev_throughput
client_latency_stddev[_vc, _mt] = client_stddev_latency
print("MW IS: ", mt)
argmax_vc = np.argmax(client_throughput_means[:, _mt].flatten())
print("Argmax vc is: ", 2**argmax_vc)
print("Max is: ", )
print("queuesize: ", queuesizes[argmax_vc])
print("waittimes: ", waittimes[argmax_vc])
print("queuetimes: ", queuetimes[argmax_vc])
print("Client troughput mean: ",
client_throughput_means[argmax_vc, _mt].flatten())
print("MW troughput mean: ", mw_throughput_means[argmax_vc,
_mt].flatten())
print(
"MW derived throughput time: ", 2 * 3 * (2**argmax_vc) /
mw_latency_means[argmax_vc, _mt].flatten() * 1000)