-
Notifications
You must be signed in to change notification settings - Fork 16
/
script_run_scaling.py
77 lines (62 loc) · 3.46 KB
/
script_run_scaling.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
from scaling.ProactiveSLAScaling import SLABasedOnVms as BrokerScaling
from utils.IOUtil import load_number_of_vms, save_scaling_results_to_csv
from utils.MetricUtil import ADI
import numpy as np
model_names = {"ann": "ann", "mlnn": "mlnn", "flnn": "flnn", "fl_gann": "fl_gann", "fl_bfonn": "fl_bfonn"}
input_types = {"uni": "uni", "multi": "multi"}
models = [
{"name": model_names["ann"],
"sliding": 3,
"input_type": input_types["uni"],
"cpu": "ANN-sliding_3-act_func1_0-act_func2_0-optimizer_0-num_hidden_20-epoch_2000-lr_0.1-batch_16",
"ram": "ANN-sliding_3-act_func1_0-act_func2_0-optimizer_0-num_hidden_20-epoch_1000-lr_0.05-batch_64"},
{"name": model_names["ann"],
"sliding": 3,
"input_type": input_types["multi"],
"cpu": "ANN-sliding_3-act_func1_0-act_func2_1-optimizer_0-num_hidden_20-epoch_1500-lr_0.05-batch_16",
"ram": "ANN-sliding_3-act_func1_0-act_func2_0-optimizer_0-num_hidden_8-epoch_1000-lr_0.1-batch_64"},
{"name": model_names["ann"],
"sliding": 4,
"input_type": input_types["uni"],
"cpu": "ANN-sliding_4-act_func1_0-act_func2_0-optimizer_0-num_hidden_20-epoch_1000-lr_0.05-batch_64",
"ram": "ANN-sliding_4-act_func1_0-act_func2_0-optimizer_2-num_hidden_20-epoch_1000-lr_0.15-batch_64"},
{"name": model_names["ann"],
"sliding": 4,
"input_type": input_types["multi"],
"cpu": "ANN-sliding_4-act_func1_0-act_func2_0-optimizer_0-num_hidden_15-epoch_1500-lr_0.15-batch_64",
"ram": "ANN-sliding_4-act_func1_0-act_func2_0-optimizer_2-num_hidden_20-epoch_1000-lr_0.15-batch_64"},
{"name": model_names["ann"],
"sliding": 5,
"input_type": input_types["uni"],
"cpu": "ANN-sliding_5-act_func1_0-act_func2_0-optimizer_0-num_hidden_20-epoch_1000-lr_0.1-batch_64",
"ram": "ANN-sliding_5-act_func1_0-act_func2_0-optimizer_3-num_hidden_15-epoch_1000-lr_0.05-batch_256"},
{"name": model_names["ann"],
"sliding": 5,
"input_type": input_types["multi"],
"cpu": "ANN-sliding_5-act_func1_0-act_func2_0-optimizer_0-num_hidden_8-epoch_2000-lr_0.05-batch_64",
"ram": "ANN-sliding_5-act_func1_0-act_func2_0-optimizer_0-num_hidden_20-epoch_2000-lr_0.15-batch_256"},
]
s_coffs = [1.0, 1.1]
L_adaps = [2, 3]
resource_real_used = load_number_of_vms('vms_real_used_CPU_RAM.csv')
for model in models:
cpu_file = "results/" + model["name"] + "/cpu/" + model["cpu"] + ".csv"
ram_file = "results/" + model["name"] + "/ram/" + model["ram"] + ".csv"
violated_arrays = []
adi_arrays = []
for s_coff in s_coffs:
violated_arr = []
adi_arr = []
for L_adap in L_adaps:
broker = BrokerScaling(scaling_coefficient=s_coff, adaptation_len=L_adap)
neural_net = broker.sla_violate(cpu_file, ram_file)
eval_scaler = ADI(metric='CPU Utilisation %')
adi = sum(np.array(eval_scaler.calculate_ADI(resource_used=resource_real_used, resource_allocated=neural_net[1][-1])))
violated_arr.append(neural_net[0])
adi_arr.append(adi)
violated_arrays.append(violated_arr)
adi_arrays.append(adi_arr)
violated_path_file = "results/scaling/sliding_" + str(model["sliding"]) + "/" + model["input_type"] + "/" + model["name"] + "_violated"
adi_path_file = "results/scaling/sliding_" + str(model["sliding"]) + "/" + model["input_type"] + "/" + model["name"] + "_adi"
save_scaling_results_to_csv(violated_arrays, violated_path_file)
save_scaling_results_to_csv(adi_arrays, adi_path_file)