Journal of Intelligent & Fuzzy Systems xx (20xx) x–xx 1

DOI:10.3233/JIFS-219200
IOS Press

1 Multi-objective task scheduling in cloud

2 computing environment by hybridized bat

f
algorithm

roo
3

4 Timea Bezdan, Miodrag Zivkovic, Nebojsa Bacanin∗ , Ivana Strumberger, Eva Tuba and Milan Tuba
Singidunum University, Danijelova, Belgrade, Serbia

rP
5

tho
6 Abstract. Cloud computing represents relatively new paradigm of utilizing remote computing resources and is becoming
7 increasingly important and popular technology, that supports on-demand (as needed) resource provisioning and releasing
8 in almost real-time. Task scheduling has a crucial role in cloud computing and it represents one of the most challenging
9 issues from this domain. Therefore, to establish more efficient resource employment, an effective and robust task allocation
(scheduling) method is required. By using an efficient task scheduling algorithm, the overall performance and service quality,
10
Au
11 as well as end-users experience can be improved. As the number of tasks increases, the problem complexity rises as well,
12 which results in a huge search space. This kind of problem belongs to the class of NP-hard optimization challenges. The
13 objective of this paper is to propose an approach that is able to find approximate (near-optimal) solution for multi-objective
14 task scheduling problem in cloud environment, and at the same time to reduce the search time. In the proposed manuscript, we
15 present a swarm-intelligence based approach, the hybridized bat algorithm, for multi-objective task scheduling. We conducted
d

16 experiments on the CloudSim toolkit using standard parallel workloads and synthetic workloads. The obtained results are
17 compared to other similar, metaheuristic-based techniques that were evaluated under the same conditions. Simulation results
cte

18 prove great potential of our proposed approach in this domain.

19 Keywords: Cloud computing, task scheduling, multi-objective optimization, bat algorithm, hybridization
rre

20 1. Introduction Task (resource) scheduling is one of the most 31

important aspect, however also a challenge in the 32

21 Cloud computing represents a relatively novel cloud computing. When end-users submit tasks (user 33

method in the information technology (IT) indus- requests) to the cloud system, they are processed
co

22 34

23 try, that delivers and manages hardware and software by a scheduling algorithm and being allocated to 35

24 as resources over the Internet. In the cloud system, the available virtual machines (VMs). The goal of 36

25 the resources are in virtual form, and that is why the task scheduling is to maximize resource utilization 37
Un

26 virtualization technology represents the main driving and to enhance the execution of tasks. Task schedul- 38

27 force of cloud computing paradigm. In the cloud plat- ing is multi-objective optimization problem and it 39

28 form, the cloud users can lease computing resources, belongs to the class of non-deterministic polynomial 40

29 such as storage, memory, CPU, applications, plat- (NP) hard challenges. Approximation algorithms are 41

30 forms for development, etc. over the network. widely utilized for solving these kind of problems. 42

For task scheduling problem in cloud computing, 43

∗ Corresponding
approximation approaches, such are metaheuristics, 44
author. Nebojsa Bacanin, Singidunum Uni-
versity, Danijelova 32, 11000 Belgrade, Serbia. E-mail:
can find optimal or a solution close to the optimum 45

[email protected]. in a reasonable time, while taking into consideration 46

ISSN 1064-1246/$35.00 © 2021 – IOS Press. All rights reserved.

 2 T. Bezdan et al. / Multi-objective task scheduling in cloud computing environment by hybridized bat algorithm

47 various performance parameters, such as completion instances and compared with other state-of-the-art 99

48 time, cost, resource utilization, and others, that all has metaheuristics. Afterwards, it was applied to practical 100

49 influence on the overall quality of service (QoS) for cloud computing scheduling problem and evaluated 101

50 the end-users. with other outstanding metaheuristics that were tested 102

51 Recently, due to its robustness, bio-inspired under the same experimental conditions. 103

52 metaheuristics have attracted attention of the The rest of the paper is organized as follows: 104

53 researchers world-wide. One of the most impor- the problem formulation is given in Section 2 and 105

54 tant representatives of nature-inspired algorithms detailed overview of the proposed hybrid metaheuris- 106

f
55 is population-based approaches known as swarm tics is shown in Section 3. Simulation results for 107

roo
56 intelligence. These methods simulate collective orga- standard unconstrained instances and practical cloud 108

57 nized behavior of group of organisms from the task scheduling problems are presented in Section 109

58 nature without any centralized coordination com- 4 and finally, Section 5 provides final remarks and 110

59 ponent. Swarm algorithms are characterized by concludes the paper. 111

60 randomization and its search process is conducted

rP
61 by two mechanisms - exploitation (intensification)
62 and exploration (diversification). In the exploration 2. Problem formulation 112

63 phase, the algorithm explores the search space glob-

64 ally, on the other hand, in the exploitation phase, the The cloud data centers hold the cloud hardware 113

tho
65 algorithm searches locally around the current best infrastructure. They have a limited amount of physi- 114

66 solutions. cal servers, typically refereed to as hosts. Each host 115

67 Many swarm algorithms are available and pre- is defined by several attributes, for example unique 116

68 sented in modern computer science literature. They identifier (hostID), number of the available process- 117

69 were successfully validated against benchmark func- ing elements (PE), performance metric for each PE 118
Au
70 tions [1–3], and also many implementations for specified in MIPS (millions of instructions per sec- 119

71 practical problems, that generate outstanding results, ond), and so on. Several VMs can be hosted on a 120

72 are available. For example, RFID network planning single physical server, either by implementing a time- 121

73 was successfully tackled with the fireworks algo- shared or a space-shared VM scheduling policy. 122

74 rithm (FWA) [4], firefly algorithm (FA), tree growth When cloud users issue requests (tasks) for pro- 123

algorithm (TGA), monarch butterfly optimization cessing, they send them to the cloud system, where the
d

75 124

76 (MBO) and artificial flora (AF) swarm algorithms task manager component receives and organizes them 125
cte

77 were successfully implemented for designing convo- and provides processing status of each task to the 126

78 lutional neural network architectures [5–8]. Swarm cloud user. After organizing the tasks, task manager 127

79 algorithms have also many other implementations forwards the tasks to task scheduler, which is respon- 128

80 such as classification and feature selection [9], wire- sible for assigning each task to the available VM 129

81 less sensor networks (WSNs) life-time optimization by utilizing the task scheduling algorithm. Described 130
rre

82 [10] and localization [11]. Moreover, according to process is shown in Figure 1. 131

83 the literature survey, many swarm algorithms imple- The VM is considered to be available if it has com- 132

84 mentations from the cloud computing domain can be pleted processing of the previously assigned tasks, 133

85 found [12–17]. and if it does not have a scheduled task ahead. The 134
co

86 The aim of research proposed in this manuscript main goal of the cloud system as a whole is to use 135

87 is to improve one instance of multi-objective task available VMs efficiently, without overloading the 136

88 scheduling in cloud computing environment by system. Fundamental goal for addressing the task 137

89 applying hybridized bat algorithm (BA) [18]. The scheduling problem in the cloud environment is to 138
Un

90 simulations are performed on standard parallel work- allocate the available resources (VMs) to the received 139

91 load traces, on the NASA Ames iPSC/860 and tasks, while achieving multiple objectives, such as 140

92 HPC2N, as well as on synthetic workloads gen- minimization of the makespan and total cost for task 141

93 erated by normal and uniform distribution. During processing. 142

94 simulations with the original BA on standard bench- The strategy of multi-objective task scheduling 143

95 mark instances, some deficiencies were observed, in a cloud computing environment, that is used in 144

96 and they are overcome in the BA’s improved version experiments, is formulated in this section. For sim- 145

97 that is proposed in this manuscript. Hybrid algorithm ulation purposes, infrastructure as a service (IaaS) 146

98 was first validated against 10 standard unconstrained cloud model is used with two objectives: financial 147
 T. Bezdan et al. / Multi-objective task scheduling in cloud computing environment by hybridized bat algorithm 3

f
roo
Fig. 1. Task scheduling in the cloud environment

148 cost reduction and the minimization of the makespan.

The same model was applied in [19].

rP
149

150 The computing resources are provided to cloud t = {t1 , t2 , t3 , . . . , tn } (4)

151 users through virtual machines (VMs). An active vir-
152 tual machine is an instance that runs the workload in where t indicates to the set of n tasks. 178

153 the cloud. The goal is to optimize the execution cost and 179

tho
154 In the proposed cloud computing model, an makespan by the metaheuristic optimization algo- 180

155 instance set can be defined as: rithm during the task allocation to the virtual instance 181

types under deadline task execution constraint. 182

The makespan is calculated according to the fol- 183

I = {I1 , I2 , I3 , ..., In } (1) lowing formula [19]: 184

Au
156 where I denotes the set of n instances.
157 Various instance series types are provided by the makespan = max{Fti : ti ∈ T } (5)
158 IaaS cloud providers, with an extensive instance type
159 range, comprising of various memory, CPU, and net- where Fti denotes the finish time of the task i. 185

160 working capacity combinations. Based on the users’ The calculation of task execution time is defined 186
d

161 computing requirements, the instances are grouped as follows: 187

162 into series. For example, Amazon EC2 currently

cte

163 offers three different instance series types: memory- si

164 intensive, compute-intensive, and storage-intensive. e(ti , vks ) = , (6)
pks
165 A type of series can be described as a set:
where the task length is denoted by si , and the com- 188
rre

puting unit is denoted by pks . The execution time

V = {V1 , V2 , V3 , . . . , Vs , . . . , VS }
189
(2)
of task i on the type of instance vks is described by 190

166 where V denotes the set of S instance types series. e(ti , vks ). 191

167 Each series type consists of instance types The IaaS’s pricing model is defined as [19]: 192

expressed by the set:

co

168

P = {P1 , P2 , P3 , . . . , Pt , . . . , Pr } (7)
Vs = {v1s , v2s , v3s , . . . , vks , . . . , vK
s } (3)
The bill function calculates the instance type of
Un

193
169 where the set of series type is denoted by Vs , consist- usage cost. 194
170 ing of K instance types (vks ). The IaaS cloud service model is defined as [19]: 195
171 The compute unit (CU) refers to the CPU capacity
172 and it is denoted by pks in an instance type (vks ). The
173 measure of CU is represented in a million floating- C = (V, Vs , P) (8)
174 point operations per second (MFLOPS), and the csk
175 indicates the cost per time unit. where V denotes the instance series type, Vs is the 196

176 The requests (tasks) submitted by the cloud end- type of instance and P represents the pricing model. 197

177 users are defined by the set: The cost calculation is formulated as [19]: 198
 4 T. Bezdan et al. / Multi-objective task scheduling in cloud computing environment by hybridized bat algorithm

where minimum and maximum frequency are fmin 228


n and fmax , respectively. Parameter β is a pseudo- 229

cost = csk × [F (ti , vks ) − S(ti , vks )], (9) random number drawn from the uniform distribution 230

i=1 in the interval [0, 1]. 231

The BA directs local search process by utilizing 232

199 where the finish time is denoted by F (ti , vks ) and the random walk, that is based on the location of current 233
200 start time is denoted as S(ti , vks ). best solution, as follows [18]: 234
201 Finally, the objective function f is calculated as:

f
xnew = xold + At ,

roo
(14)
f = (makespan, cost)T (10)
where At indicates to the mean value of all individ- 235

uals’ loudness, and it is scaled by the  parameter, 236

that is drawn from the uniform distribution within 237

202 3. Details of proposed hybridized BA the range [−1, 1].

rP
238

The bats’ pulse emission loudness update occurs 239

203 In this section, after brief description of the original when a bat finds the prey by using the following 240
204 BA and its deficiencies, the hybridized BA version, expression: 241
205 that overcomes drawbacks of original implementa-
tion, is presented in detail.

tho
206

Ati = αAt−1
i , ri = ri [1 − exp(−γt)]
t 0
(15)
207 3.1. Original BA

208 The BA is a very efficient nature-inspired, Ati → 0, rit → ri0 , while t → ∞ (16)
Au
209 population-based optimization algorithm developed
210 in 2010 by Xin-She Yang [18]. The algorithm is where Ati and r t denote loudness and pulse emission 242

211 inspired by the bats’ echolocation behavior. The bats rate of i-th individual at iteration t, respectively. The 243

212 are searching for the prey by emitting ultrasonic α and γ are constants, which typically take value 244

213 sound waves. The reflected echo from the objects between 0 and 1. 245

helps them to sense the distance, as well as to dif- At the beginning of algorithm’s execution initial
d

214 246

215 ferentiate between preys, foods, and other types of values for loudness (A0i ) and pulse emission rate (ri0 ) 247

objects. are set for all solutions, and as the iterations progress,
cte

216 248

217 Individuals (bats) in the population update their these values will be updated for only those individu- 249

218 position and velocity, with the increase in iteration als that are improved in terms of convergence towards 250

219 number t as follows [18]: optimum solution. Optimal choice of control param- 251

eters’ values should be determined for each particular 252

rre

problem by conducting simulations. 253

xit = xit−1 + vti , (11)
3.2. Deficiencies of basic BA and proposed 254

improvements 255
vti = vt−1 + (xit − x∗ ) · fi , (12)
co

i
The BA has very strong exploitation abil-
where xit−1 indicates current location of individual
256
220
ity, however extensive practical simulations show 257
221 i, xit is its new (updated) position and vt−1i and vti that its exploration capability and intensification- 258
Un

222 denote current and new individual’s velocity, respec- diversification balance could be enhanced [20]. The 259
223 tively. The current best solution in the population is BA’s selection process is relatively stable during the 260
224 represented by x∗ , while the frequency of i-th solution run by directing individuals towards the current best 261
225 is denoted by fi . solution ((Eq. 12)). Orientation towards the current 262
226 The bat’s frequency is calculated by the following best typically establishes good results in later itera- 263
227 formula: tions, when the search process has converged to an 264

optimum region. However, in early iterations, if ini- 265

fi = fmin + (fmax − fmin ) · β, (13) tially generated random solutions are far from the 266

optimum region, such orientation may lead to the pre- 267

 T. Bezdan et al. / Multi-objective task scheduling in cloud computing environment by hybridized bat algorithm 5

268 mature convergence and the search process may stuck where lbj and ubj denote lower and upper bounds 312

in sub-optimal domains. lb +ub

269
of component j, respectively, j 2 j represents 313
270 This problem can be also observed from the aspect arithmetic mean of the interval [lbj , ubj ], while 314
271 of intensification-diversification trade-off, which is lb +ub
rnd( j 2 j , xj) generates uniform random value 315
272 adjusted in favor to intensification. Moreover, some lb +ub
273 research show that observed original BA’s deficien- from the interval [ j 2 j , xj ]. 316

274 cies are more emphasized in problems with larger Inspired by proposed modifications, hybridized 317

275 dimensions [21]. In such cases BA’s drawbacks can BA approach is named BA ABC exploitation quasi- 318

f
276 be addressed by balancing with the search equation of reflection learning (BAAEQRL). 319

roo
277 some other method, that is not oriented towards the
278 current best and/or by utilizing explicit exploration 3.3. The BAAEQRL workings and pseudo-code 320

279 mechanism.
280 Method proposed in this manuscript addresses At the beginning of execution, initial population P 321

281 BA’s deficiencies by introducing two modifica- with dimension N × M, that consists of N individ- 322

rP
282 tions. First, original BA’s not appropriately adjusted uals (xi , i = 1, ...N), is created within a predefined 323

283 exploitation-exploration balance is addressed by lower (lbj ) and upper (ubj ) (j = 1, ...M) parameters’ 324

284 incorporating employee bee search procedure, that bounds randomly: 325

285 conducts intensification process, from the well-

tho
286 known ABC metaheuristics [1]. Second, at the
287 end of each iteration, quasi-reflection-based learning xi,j = lbj + rand(ubj − lbj ), (19)
288 (QRBL) mechanism is triggered, that improves both
where the random uniform number is denoted with 326
289 - the exploration ability and achieves better trade-off
rand and xi,j indicates to the j-th component of i-th 327
290 between intensification and diversification.
individual.
Au 328
291 Approach that is shown in this manuscript adopts
After the initial population is created, each solu- 329
292 ABC’s exploitation procedure as proposed in [22]:
tion is evaluated for fitness, by using the following 330

expression in case of minimization problems: 331

xit = xit−1 +φ · (xit−1 − xkt−1 ), (17)

1
if oi ≥ 0
d

293 where xit and xit−1 , denote new and previous location fiti = Fi
(20)
294 of individual xi at time steps t and t − 1, respectively, 1 + |Fi | otherwise,
cte

295 xkt−1 is location of the k random individual from the

296 population at time step t − 1, while φ represents num- where fiti denotes the fitness of i-th solution, and the 332

297 ber drawn from the uniform distribution within the objective function of the i-th solution is represented 333

298 interval [−1, 1]. by Fi . 334

As noted above, in each even iteration (t%2 == 0)

rre

299 It should be noted that the modification rate (MR) 335

300 parameter is not used as in [22]. BA’s search equation is triggered, while in each odd 336

301 The balance between BA’s and ABC search proce- iteration (t%2! = 0) the search process is executed 337

302 dures is established in the following way: the ABC by the ABC search procedure, as shown in Eq. (17). 338

exploitation is executed in each odd iteration, while At the end of each iteration t, the QRBL is applied 339
co

303

304 the basic BA’s search is triggered in each even itera- (Eq. (18)) to determine quasi-reflective population: 340

305 tions.
306 As noted above, a second modification (QRBL) qr
P qr = {Xi,j } (21)
Un

307 is also introduced in the basic BA approach. By

308 incorporating QRBL mechanism in meataheuristics where i = 1, 2, 3..., N, and j = 1, 2, 3, ...M of cur- 341
309 better convergence rate and solutions diversity can be rent population P. 342
310 achieved [23]. Quasi-reflected parameter j of solution Afterwards, the original population P and popu- 343
311 x is calculated in the following way: lation P qr are merged (P ∪ P qr ) and individuals in 344

merged population are sorted in descending order 345

qr lbj + ubj according to its fitness value. Finally, N best solu- 346
xj = rnd( , xj ), (18) tions are selected as the new population for the next
2 347

iteration t + 1. 348
 6 T. Bezdan et al. / Multi-objective task scheduling in cloud computing environment by hybridized bat algorithm

349 As can be seen from the BAAEQRL details, pro- objective cloud scheduling problem are presented 371

350 posed approach does not utilize additional control along with comparative analysis with other state-of- 372

351 parameters, however, in each iteration, due to the the-art methods. 373

352 QRBL procedure, BAAEQRL performs 2 · N func-

353 tion evaluations, while the original BA conducts 4.1. Unconstrained benchmark simulations 374

354 only N evaluations. For that reason to establish

355 fair comparative analysis with the original BA, the Proposed BAAEQRL approach was validated 375

356 BAAEQRL should be tested with fewer number of on a 10 classical benchmark instances. Moreover, 376

f
357 iterations. extensive comparative analysis with state-of-the-art 377

roo
358 The proposed method’s pseudo-code is provided approaches presented in [21], as well as with the 378

359 in the Algorithm 1. original BA, was performed. Details of benchmark 379

instances utilized in simulations are shown in Table 1. 380

All functions were tested with 30 dimensions (D = 381

Algorithm 1 Pseudo-code of proposed BAAEQRL
30). 382

rP
Define objective function F (x)
Initialize random initial population according to Eq. (19) The following metaheuristics were included in 383
For each solution i define the values of parameters vi , ri , Ai , and the comparative analysis: original BA, directional BA 384
frequency of pulse (fi ) at xi
Set the iteration counter t to 0 and define maximum number of iterations (dBA), particle swarm optimization (PSO), harmony 385
(MaxIter) search (HS), cuckoo search (CS), genetic algorithms 386
Evaluate fitness of each solution

tho
while t < MaxIter do (GA) and differential evolution (DE). The dBA as 387
for i = 1 to N do state-of-the-art approach was presented in [21], and 388
if t is even then
Calculate the velocity and frequency value by using Eq. (12) also, the results of all other approaches included in 389
and Eq. (13), respectively analysis were retrieved form this paper. We note that 390
Perform the BA search procedure using Eq. (11)
if rand > ri then for the purpose of this research we have also per- 391
Select the best solution
Au
formed experiments with original BA and obtained 392
Perform the random walk process by using Eq. (14)
end if similar results as in [21]. 393
Randomly generate new solution All algorithms taken for comparative analysis were 394
if (pi < Ai and f (xi ) < f (x∗ ) then
The newly generated solution is accepted tested with 15.000 function evaluations excluding 395
Reduce Ai and increase ri by utilizing Eq. (15) initialization phase and with 30 solutions in popu- 396
end if
lation, which yields in total number of 500 iterations
d

else 397
Perform the ABC search procedure by using Eq. (17) (15.000/30), as in [21]. Due to the fact that pro- 398
if f (xi ) < f (x∗ ) then
posed BAAEQRL in each iterations evaluates 2 · N
cte

The newly generated solution is accepted 399

end if solutions, it was tested with only 250 iterations 400
end if
end for (15.000/2 · 30). 401
Generate population P qr , merge populations P and P qr , sort all Standard BA and proposed BAAEQRL were 402
individuals according to fitness and select N best solutions
Find and save the current best solution x∗ tested with the following control parameters’ adjust- 403
rre

end while ments: r0 = 0.1, A0 = 0.9, α = γ = 0.9, fmin = 0 404

Return the best solution
Post-processing and visualization and fmax = 2. By conducting empirical simulations 405

with various parameters’ values, we have determined 406

that with these set of control parameters in average 407

co

BA and BAAEQRL establish the best performance. 408

360 4. Experimental results Control parameters’ adjustments of other methods 409

can be retrieved from [21]. 410

361 Before validating proposed method on practi- Simulation results are presented in Table 2, where 411
Un

362 cal challenge of multi-objective task scheduling in the best result for each metric is marked bold. All 412

363 cloud environment, following good practice from metrics - best, median, worst, average and standard 413

364 recent computer science literature, simulations are deviation (SD) are calculated based on 30 indepen- 414

365 performed on a well-known group of global uncon- dent runs. 415

366 strained benchmarks. For that reason, in the first part Comparative analysis showed in Table 2 proves 416

367 of this section, simulations on standard 10 uncon- that in average, for all benchmark instances, pro- 417

368 strained test instances, along with parameter setup posed BAAEQRL establishes better results quality, 418

369 and comparative analysis, are shown. Afterwards, as well as convergence speed than other state-of-the- 419

370 results of empirical simulations for practical multi- art metaheuristics that were taken for comparison. 420
 T. Bezdan et al. / Multi-objective task scheduling in cloud computing environment by hybridized bat algorithm
Table 1

ID Name
Un Formulation
d
Unconstrained benchmark function details used in simulations
Search Range Optimum Parameters
f1
f2
Sphere
Sum of Different powers
co
f (x) =
f (x) =
 i=1 i
d
i=1 i
x2
|x |i+1
[−100, 100]d
[−100, 100]d
0
0
x∗ = (0, ..., 0)
x∗ = (0, ..., 0)

rre
 d  i
f3 Rotated hyper-ellipsoid f (x) = i=1
x2
j=1 j
[−65, 65]d 0 x∗ = (0, ..., 0)
 d xi2  d
f4 Griewank f (x) = − i=1 cos( √ xi
)+1 [−600, 600]d 0 x∗ = (0, ..., 0)

cte
i=1 4000
 d d
i
f5 Trid f (x) = (x − 1) −
i=1 i
2 xx
i=2 i i−1
[−d 2 , d 2 ]d −d(d + 4)(d − 1)/6 xi = i(d + 1 − i)
 d
f6 Rastrigin f (x) = 10d + [x − 10 cos(2πxi )]
2 [−5.12, 5.12]d 0 x∗ = (0, ..., 0)

d
i=1 i
 d−1
f7 Levy f (x) = sin2 (πw1 ) + i=1
(wi − 1)2 [1 + 10sin2 (πwi + 1)]+ [−5.12, 5.12]d 0 x∗ = (1, ..., 1)

f8 Ackley
+(wd − 1)2 [1 + 10sin2 (πw

f (x) = −a × exp(−b
 
1 d
d
d )], where

x2 ) − exp( d1
i=1 i
d
Au
: wi = 1 + (xi − 1)/4

cos(cxi )) + a + exp(1) [−32, 32]d 0 x∗ = (0, ..., 0)

tho
i=1
where a = 20, b = 0.2
d √
f9 Schwefel f (x) = 418.9829d × d − x
i=1 i
sin( |xi |) [−500, 500]d 0 x∗ = (420.9687, ..., 420.9687)
d−1
f10 Rosenbrock f (x) = i=1
(100(xi2 − xi+1 )2 + (1 − xi )2 )

rP [−10, 10]d 0 x∗ = (1, ..., 1)

roo
f

7
 8
Table 2
Comparative analysis for 30-dimensional benchmark functions
Function dBA BA PSO HS CS GA DE BAAEQRL
f1 Best 1.927E-03 3.052E-01 1.118E+03 5.919E+03 2.340E+02 5.517E+00 2.481E+01 7.458E-08
Median 1.408E-02 5.480E+04 2.554E+03 9.621E+03 4.357E+02 6.560E+02 4.120E+01 4.703E-06

T. Bezdan et al. / Multi-objective task scheduling in cloud computing environment by hybridized bat algorithm
Worst 2.233E+00 6.569E+04 5.626E+03 1.568E+04 6.119E+02 7.964E+03 8.028E+01 5.292E-01
Mean 2.256E-01 4.920E+04 2.852E+03 9.618E+03 4.153E+02 1.678E+03 4.411E+01 9.841E-05
SD 4.869E-01 1.859E+04 1.105E+03 2.226E+03 9.518E+01 2.032E+03 1.259E+01 4.730E-04
f2 Best 1.011E+06 3.313E+09 1.609E+20 2.573E+33 3.229E+17 7.488E+04 9.080E+08 7.076E-03
Median 8.171E+09 1.294E+45 1.085E+28 7.580E+37 7.654E+19 9.245E+29 1.177E+11 6.581E+01
Worst 1.713E+13 5.893E+50 1.724E+34 8.664E+42 2.433E+22 2.390E+41 1.553E+12 5.066E+04
Mean 1.363E+12 4.310E+49 1.046E+33 3.533E+41 2.263E+21 1.049E+40 3.051E+11 5.458E+02
SD 4.261E+12 1.461E+50 3.737E+33 1.697E+42 5.976E+21 4.671E+40 4.102E+11 4.713E+01

Un
f3 Best 1.634E-02 8.563E+00 4.828E+03 4.124E+04 1.062E+03 8.280E+01 9.877E+01 3.256E-03
Median 3.115E-01 2.996E+05 1.383E+04 5.220E+04 1.996E+03 5.373E+03 1.618E+02 7.546E-01
Worst 1.256E+02 4.370E+05 3.416E+04 7.472E+04 3.409E+03 3.294E+04 3.850E+02 1.343E+01
Mean 1.461E+01 2.612E+05 1.562E+04 5.336E+04 2.138E+03 8.130E+03 1.742E+02 9.135E+00

co
SD 3.456E+01 1.348E+05 7.676E+03 8.132E+03 5.493E+02 8.472E+03 6.173E+01 8.332E+01
f4 Best 5.049E-03 3.210E+02 3.041E+01 4.375E+01 3.026E+00 1.080E-01 9.989E-03 8.250E-06
Median 8.544E-02 5.949E+02 7.258E+01 8.306E+01 4.448E+00 1.507E+01 8.997E-02 7.029E-04

rre
Worst 5.630E-01 6.848E+02 1.684E+02 1.201E+02 6.797E+00 5.574E+01 2.136E+00 2.752E-02
Mean 1.405E-01 5.816E+02 7.481E+01 8.040E+01 4.567E+00 1.900E+01 2.303E-01 2.315E-03
SD 1.481E-01 7.884E+01 2.717E+01 1.588E+01 9.934E-01 1.828E+01 4.210E-01 2.085E-01
f5 Best 1.685E+03 2.967E+06 3.078E+05 5.169E+05 2.831E+04 6.326E+03 -3.276E+03 -4.857E+03
Median
Worst
Mean
SD
3.553E+04
9.707E+04
3.423E+04
2.590E+04 cte
4.529E+06
5.495E+06
4.436E+06
6.360E+05
5.827E+05
1.223E+06
6.204E+05
2.312E+05
8.395E+05
1.329E+06
8.815E+05
1.932E+05
4.084E+04
8.620E+04
4.242E+04
1.118E+04
2.920E+05
7.001E+05
3.194E+05
1.916E+05
3.007E+03
2.215E+04
4.901E+03
6.627E+03
-2.559E+03
-1.549E+03
-2.105E+03
9.521E+01
f6

f7
Best
Median
Worst
Mean
SD
Best
6.812E+01
1.057E+02
2.471E+02
1.193E+02
4.023E+01
1.518E+00
2.420E+02
3.074E+02
3.670E+02
3.086E+02
3.603E+01
3.024E+01
d 1.707E+02
2.517E+02
3.456E+02
2.599E+02
3.756E+01
2.126E+01 Au
1.330E+02
1.625E+02
1.845E+02
1.580E+02
1.558E+01
1.366E+01
1.129E+02
1.378E+02
1.644E+02
1.366E+02
1.349E+01
2.414E+00
2.994E+01
5.895E+01
9.913E+01
5.746E+01
1.825E+01
1.093E+00
2.998E+01
1.575E+02
2.047E+02
1.551E+02
3.368E+01
1.053E+00
3.852E-01
6.872E+01
1.537E+02
3.702E+01
1.905E+01
4.328E-01
Median
Worst
Mean
SD
4.901E+00
9.997E+00
4.716E+00
1.826E+00
6.876E+01
1.135E+02
7.176E+01
1.927E+01
3.604E+01
8.057E+01
3.979E+01
1.681E+01 tho2.384E+01
3.540E+01
2.417E+01
5.004E+00
4.475E+00
8.813E+00
5.153E+00
1.865E+00
4.073E+00
1.562E+01
5.675E+00
3.920E+00
1.928E+00
3.388E+00
2.017E+00
5.223E-01
1.815E+00
3.721E+00
1.992E+00
0.525E+00

rP
f8 Best 3.214E+00 1.996E+01 1.252E+01 1.338E+01 8.691E+00 2.595E+00 2.302E+00 5.543E-09
Median 5.681E+00 1.996E+01 1.462E+01 1.559E+01 1.200E+01 5.744E+00 3.191E+00 1.482E-07
Worst 8.801E+00 1.996E+01 1.737E+01 1.640E+01 1.750E+01 1.145E+01 3.648E+00 6.766E-04
Mean 5.839E+00 1.996E+01 1.474E+01 1.540E+01 1.209E+01 5.920E+00 3.191E+00 3.292E-06

f9
SD
Best
Median
Worst
1.730E+00
2.895E+03
4.492E+03
5.646E+03
7.062E-04
5.685E+03
9.365E+03
1.017E+04
1.235E+00
7.293E+03
8.803E+03
9.480E+03
7.839E-01
2.281E+03
3.698E+03
4.624E+03
1.753E+00
4.522E+03
5.045E+03
5.426E+03 roo
2.453E+00
2.736E+03
4.228E+03
5.993E+03
2.904E-01
4.745E+03
5.370E+03
6.006E+03
2.335E-05
2.941E+02
2.941E+02
2.941E+02

f
Mean 4.357E+03 8.940E+03 8.712E+03 3.722E+03 5.056E+03 4.208E+03 5.407E+03 2.941E+02
SD 6.414E+02 1.242E+03 5.463E+02 5.060E+02 1.747E+02 7.320E+02 3.363E+02 0.000E+00
f10 Best 2.911E+01 3.336E+01 8.566E+03 8.437E+04 6.691E+02 1.048E+02 4.637E+02 0.052E+00
Median 1.038E+02 2.473E+02 5.394E+04 1.588E+05 9.105E+02 2.756E+03 6.892E+02 0.052E+00
Worst 1.011E+03 2.944E+03 2.811E+05 2.346E+05 2.290E+03 4.793E+04 1.304E+03 2.871E+02
Mean 1.645E+02 4.916E+02 8.159E+04 1.597E+05 1.073E+03 5.961E+03 7.193E+02 2.045E+00
SD 1.926E+02 6.275E+02 6.481E+04 4.048E+04 3.967E+02 9.588E+03 2.121E+02 7.143E+00
 T. Bezdan et al. / Multi-objective task scheduling in cloud computing environment by hybridized bat algorithm 9

Table 3
Statistical comparison between the BAAEQRL and other approaches with Wilcoxon Signed-Rank Test (α = 0.05)
Function BAAEQRL dBA BA PSO HS CS GA DE
f1 9.841E-05 2.256E-01 4.920E+04 2.852E+03 9.618E+03 4.153E+02 1.678E+03 4.411E+01
f2 5.458E+02 1.363E+12 4.310E+49 1.046E+33 3.533E+41 3.533E+41 1.049E+40 3.051E+11
f3 9.135E+00 1.461E+01 2.612E+05 1.562E+04 5.336E+04 2.138E+03 8.130E+03 1.742E+02
f4 2.315E-03 1.405E-01 5.816E+02 7.481E+01 8.040E+01 4.567E+00 1.900E+01 2.303E-01
f5 -2.105E+03 3.423E+04 4.436E+06 6.204E+05 8.815E+05 4.242E+04 3.194E+05 4.901E+03
f6 3.702E+01 1.193E+02 3.086E+02 2.599E+02 1.580E+02 1.366E+02 5.746E+01 1.551E+02

f
f7 1.992E+00 4.716E+00 7.176E+01 3.979E+01 2.417E+01 5.153E+00 5.675E+00 2.017E+00

roo
f8 3.292E-06 5.839E+00 1.996E+01 1.474E+01 1.540E+01 1.209E+01 5.920E+00 3.191E+00
f9 2.941E+02 4.357E+03 8.940E+03 8.712E+03 3.722E+03 5.056E+03 4.208E+03 5.407E+03
f10 2.045E+00 1.645E+02 4.916E+02 8.159E+04 1.597E+05 1.073E+03 5.961E+03 7.193E+02
p-value 9.77E-04 9.77E-04 9.77E-04 9.77E-04 9.77E-04 9.77E-04 9.77E-04

rP
421 The most significant difference can be noticed in speed improvements over the original BA. For 459

422 f 1, f 2, f 8, f 9 and f 10 benchmarks and in these that purpose we implemented BA with only ABC 460

423 tests the BAAEQRL obtained better results that all exploitation (BAAE) and BA with only QRBL mech- 461

424 other approaches for all metrics (best, median, worst, anism (BAQRL) and generated convergence speed 462

meand and standard deviation). graphs for all four approaches. It should be noted that

tho
425 463

426 State-of-the-art dBA, proposed in [21], established since the BAQRL also utilizes QRBL mechanism, it 464

427 better performance than our BAAEQRL only for was also tested with only 250 iterations. Convergence 465

428 median and SD metrics in f 3 benchmark and SD speed graphs are shown in Figure 3. 466

429 metrics for f 4 benchmark. In the case of simulations

for Rastrigin function (f 6) and Levy (f 7), GA, DE
Au
430 4.2. Task scheduling in cloud environment 467

431 and CS obtained better values for only few metrics simulations 468

432 than BAAEQRL.

433 Furthermore, to determine whether improvements The CloudSim toolkit is utilized to conduct the 469

434 of proposed BAAEQRL over other approaches for experiments for multi-objective task scheduling by 470

unconstrained instances are statistically significant, the proposed BAAEQRL metaheuristics. In this
d

435 471

436 we applied Wilcoxon Signed Rank-Test to make the work, the simulation is conducted on one instance 472

pair-wise comparison between the proposed BAAE- type and one pricing option. The simulation and sys-
cte

437 473

438 QRL and other metaheuristics. In the statistical tem model is set based on the work in [19] and 474

439 analysis, we included all benchmark function, which described in Section 2. The following cloud infras- 475

440 represents the independent variables, and the depen- tructure was used in experiments: 476

441 dent variable represents the average value of each

− single datacenter with two hosts,
rre

477
442 algorithms and functions.
443 Results of Wilcoxon test are summarizes in − 1 TB storage capacity, 478

444 Table 3. The p-value obtained in the test is in all − VM instance with 2048 MB RAM, 479

445 cases < 0.05 which indicates to significant difference − 10 Gbps bandwidth, 480

between the proposed algorithm and all other com- − Xen VMM, 481
co

446

447 pared methods. Since the proposed method resulted − Linux operating system, 482

448 in a better mean value over all other metaheuristics, − x86 architecture. 483

449 the sign is "-" for all functions in each pair difference The number of VMs was set to 20. The task length 484
Un

450 observation, and that yields to the same p-value in all is in the range between 5000 GB and 50000 GB, the 485
451 pair tests. size of file ranges between 10 GB and 100 GB, and the 486
452 To better visualize search process of BAAEQRL, memory is between 10 GB and 100 GB. The virtual 487
453 we plotted 2D Gaussian Kernel and surface plots for machine types and configuration, along with the pric- 488
454 some functions using 100 iterations. Visual represen- ing, are presented in Table 4. The control parameters 489
455 tation is provided in Figure 2. of the scheduler algorithms are depicted in Table 5, 490
456 Furthermore, we wanted to determine influence and the workload settings in Table 6. 491
457 of ABC exploitation and QRBL on the BAAE- We evaluated the effectiveness of proposed method 492
458 QRL performance and also to visualize convergence on the widely used and well-known benchmarks for 493
 10 T. Bezdan et al. / Multi-objective task scheduling in cloud computing environment by hybridized bat algorithm

f
roo
rP
tho
Fig. 2. 2D Gaussian Kernel and Surface plots for some benchmarks of proposed BAAEQRL
d Au
cte
rre
co
Un

Fig. 3. Convergence speed of some benchmarks for BA, BAAE, BAQRL and BAAEQRL

494 performance evaluation in a distributed system, on weight coefficient of 0.5, while in the referred paper, 501

495 the NASA Ames iPSC/860 and HPC2N set log, as the Pareto optimality concept was utilized. Between 502

496 well as on synthetic workloads generated by nor- tasks, does not exist any precedence constraint and 503

497 mal and uniform distribution. As metrics, we utilized their executions are non-preemptive. In order to 504

498 the cost, makespan, and the Hyervolume indica- obtain statistically meaningful results, we repeated 505

499 tor. In the proposed method, for the objectives, the the algorithm testing 30 times. Performance of the 506

500 weighted sum technique is used and we set an equal BAAEQRL is compared to similar task scheduling 507
 T. Bezdan et al. / Multi-objective task scheduling in cloud computing environment by hybridized bat algorithm 11

Table 4
VM type and configuration
Name vCPU SSD Storage Memory Processing capacity Cost/hour
(GB) (GB) (MFLOPS) ($)
c3.large 2 2x16 3.75 8800 0.105
c3.xlarge 4 2x40 7.5 17600 0.210
c3.2xlarge 8 2x80 15 35200 0.420
c3.4xlarge 16 2x160 30 70400 0.840
c3.8xlarge 32 2x320 60 140800 1.680

f
roo
Table 5 EMS-C is 7.5%-9%, and over CMSOS the perfor- 532
Hybridized bat algorithm control parameters mance improvement is between 3% and 4.8%. 533

Parameter Notation Value On the synthetic workloads, the percentage 534

Population size N 20 improvement is between 1% and 23%. The perfor- 535

Initial pulse emission rate r0 0.1 mance enhancements of the proposed BAAEQRL 536

rP
Maximum initial loudness A0 0.9
Constant minimum loudness Amin 1 over BOGA ranges between 9% and 23%, on 537

Maximum frequency fmax 2 the uniform workloads, while in case of the 538

Minimum frequency fmin 0 random workloads, the improvement is between 539

Constant parameter α 0.9
16% and 20%. BAAEQRL’s results comparing to 540
Constant parameter γ 0.9

tho
ECMSMOO, on the uniform workloads is bet- 541

ter for 6%-14%, while on the random workloads, 542

Table 6 BAAEQRL has an improvement between 12%-14%. 543

Settings of the workloads BAAEQRL performance over EMS-C on both syn- 544

Parameter Value thetic workloads are between 4%-9%, while over 545
Au
Length [5000, 50 000] MFLOPS the CMSOS, BAAEQRL’s performance on both syn- 546

Memory [10, 100] GB thetic workloads are between 1%-5%. 547

File size [10, 100] GB Fig. 5 depicts the relationship between the cost and 548

makespan, where the hybrid BA scheduler shows a 549

constantly higher performance than other techniques. 550

methods, where also metaheuristic algorithms were
d

508 The proposed BAAEQRL approach outperformed 551

509 utilized (EMS-C, ECMSMOO, BOGA, and CMSOS) EMS-C, ECMSMOO, BOGA, CMSOS and the orig- 552
cte

510 and their results are taken from [19]. inal BA on all standard and synthetic workloads 553
511 Hypervolume improvement is presented in Fig. 4. instances. 554
512 By observing the graph, we can draw a conclusion
513 that the proposed hybrid BA outperformed all other
514 counterparts with different task sizes on all log sets
rre

515 NASA Ames iPSC/860, HPC2N, Random and Uni- 5. Conclusion 555

516 form. The original BA has a good performance across

517 the workload, however, the proposed improved algo- Task scheduling is a challenging problem in the 556

518 rithm shows significant performance improvement cloud computing model, due to the direct influence 557
co

519 over all compared approaches. on the performance. To resolve this particular issue, in 558

520 The performance improvement of the proposed this work, hybridized BA (BAAEQRL), task sched- 559

521 BAAEQRL on the NASA workload over BOGA uler algorithm is proposed. The cloud system model 560

522 ranges between 3% and 17%, over the ECMSMOO used in experiments represents a multi-objective opti- 561
Un

523 approach the performance improvement is between mization problem. The financial cost reduction and 562

524 10% and 14%, in the comparison with EMS-C the minimization of the makespan objectives were 563

525 method, BAAEQRL has an improvement between used in objective function formulation. 564

526 6% and 9%, while over CMSOS, the performance Proposed BAAEQRL was firstly tested on 10 565

527 improvement is between 2% and 4.7%. standard unconstrained benchmark instances and 566

528 On the HPC2N workload comparison, BAAEQRL obtained better results than other state-of-the-art 567

529 performance improvement ranges between 3% and approaches. In order to evaluate performance of pro- 568

530 22%. BAAEQRL improvement over BOGA ranging posed method for multi-objective task scheduling 569

531 8%-22%, over ECMSMOO is from 11% to 14%, over challenge, simulations are performed on standard par- 570
 12 T. Bezdan et al. / Multi-objective task scheduling in cloud computing environment by hybridized bat algorithm

f
roo
rP
tho
Fig. 4. Hypervolume improvemnt of the proposed method
Au
d
cte
rre
co
Un

Fig. 5. Relationship of the cost and makespan

571 allel workload traces, on the NASA Ames iPSC/860 Contributions of proposed manuscript are twofold: 578

572 and HPC2N, as well as on synthetic workloads gener- first the basic BA is improved by hybridization 579

573 ated by normal and uniform distribution. Similarly as with ABC metaheuristics and by introducing QRBL 580

574 in tests with standard benchmark instances, compar- mechanism and secondly, multi-objective cloud 581

575 ative analysis was performed with other outstanding task scheduling problem is addressed more effi- 582

576 algorithms and proposed BAAEQRL managed to ciently that previous methods shown in the modern 583

577 obtain better cost reduction and the makespan. literature. 584
 T. Bezdan et al. / Multi-objective task scheduling in cloud computing environment by hybridized bat algorithm 13

585 In future work, we plan to incorporate more objec- [11] I. Strumberger, M. Minovic, M. Tuba and N. Bacanin, Per- 638

586 tives in the task scheduling cloud system model, to formance of elephant herding optimization and tree growth 639
algorithm adapted for node localization in wireless sensor 640
587 make it more realistic, as well as to implement and networks, Sensors 19(11) (2019), 2515. 641
588 to improve other swarm intelligence algorithms for [12] M. Kalra and S. Singh, A review of metaheuristic schedul- 642

589 tackling this very important challenge. ing techniques in cloud computing, Egyptian Informatics 643
Journal 16(3) (2015), 275–295. ISSN 1110-8665. doi: 644
https://doi.org/10.1016/j.eij.2015.07.001. 645
[13] K. Sreenu and M. Sreelatha, W-scheduler: whale opti- 646
590 Acknowledgment mization for task scheduling in cloud computing, Cluster 647

f
Computing (2017). ISSN 1573-7543. doi: 10.1007/s10586- 648

roo
591 The paper is supported by the Ministry of Edu- 017-1055-5 649
[14] N. Bacanin, E. Tuba, T. Bezdan, I. Strumberger and M. Tuba, 650
592 cation, Science and Technological Development of Artificial flora optimization algorithm for task scheduling in 651
593 Republic of Serbia, Grant No. III-44006. cloud computing environment, In International Conference 652
on Intelligent Data Engineering and Automated Learning, 653
pages 437–445. Springer, (2019). 654

rP
[15] I. Strumberger, M. Tuba, N. Bacanin and E. Tuba, Cloudlet 655
594 References scheduling by hybridized monarch butterfly optimization 656
algorithm, Journal of Sensor and Actuator Networks 8(3) 657

595 [1] D. Karaboga and B. Akay, A modified artificial bee colony (2019), 44. doi: https://doi.org/10.3390/jsan8030044. 658

596 (ABC) algorithm for constrained optimization problems, [16] I. Strumberger, E. Tuba, N. Bacanin and M. Tuba, 659

597 Applied Soft Computing 11(3) (2011), 3021–3031. Dynamic tree growth algorithm for load scheduling 660

tho
598 [2] N. Bacanin and M. Tuba, Artificial bee colony (ABC) algo- in cloud environments, In 2019 IEEE Congress on 661

599 rithm for constrained optimization improved with genetic Evolutionary Computation (CEC) (2019), 65–72. doi: 662

600 operators, Studies in Informatics and Control 21(2) (2012), 10.1109/CEC.2019.8790014 663

601 137–146. [17] N. Bacanin, T. Bezdan, E. Tuba, I. Strumberger, M. Tuba 664

602 [3] M. Tuba and Nebojsa, Improved seeker optimization and M. Zivkovic, Task scheduling in cloud computing 665

603 algorithm hybridized with firefly algorithm for con- environment by grey wolf optimizer, In 2019 27th Telecom- 666
Au
604 strained optimization problems, Neurocomputing 143 munications Forum (TELFOR) 1–4. IEEE, (2019). 667

605 (2014), 197–207. doi: 10.1016/j.neucom.2014.06.006 [18] [18] X.-S. Yang, A New Metaheuristic Bat-Inspired 668

606 [4] I. Strumberger, E. Tuba, N. Bacanin, M. Beko and M. Algorithm, pages 65–74. Springer Berlin Heidelberg, 669

607 Tuba, Bare bones fireworks algorithm for the rfid net- Berlin, Heidelberg, (2010). ISBN 978-3-642-12538-6. doi: 670

608 work planning problem, In 2018 IEEE Congress on 10.1007/978-3-642-12538-6 6 671

609 Evolution ary Computation (CEC), (2018), 1–8. doi: [19] M. Abdullahi, M.A. Ngadi, S.I. Dishing, S.M. Abdul- 672
hamid and B.I. Ahmad, An efficient symbiotic organisms 673
d

610 10.1109/CEC.2018.8477990
611 [5] L. Cheng, X.-han Wu and Y. Wang, Artificial flora (AF) search algorithm with chaotic optimization strategy for 674

612 optimization algorithm, Applied Sciences 8 (2018), 329. doi: multi-objective task scheduling problems in cloud com- 675
cte

613 10.3390/app8030329 puting environment, Journal of Network and Computer 676

614 [6] T. Bezdan, E. Tuba, I. Strumberger, N. Bacanin and M. Applications 133 (2019), 60–74. ISSN 1084-8045. doi: 677

615 Tuba, Automatically designing convolutional neural net- https://doi.org/10.1016/j.jnca.2019.02.005. 678

616 work architecture with artificial flora algorithm, In ICT [20] M. Tuba and N. Bacanin. Hybridized bat algorithm 679

617 Systems and Sustainability 371–378. Springer, (2020). for multi-objective radio frequency identification (rfid) 680

618 [7] N. Bacanin, T. Bezdan, E. Tuba, I. Strumberger and M. Tuba, network planning. In 2015 IEEE Congress on Evo- 681
rre

619 Optimizing convolutional neural network hyperparameters lutionary Computation (CEC) (2015), 499–506. doi: 682

620 by enhanced swarm intelligence metaheuristics, Algorithms 10.1109/CEC.2015.7256931 683

621 13(3) (2020), 67. [21] A. Chakri, R. Khelif, M. Benouaret and X.-S. 684

622 [8] N. Bacanin, T. Bezdan, E. Tuba, I. Strumberger and M. Tuba, Yang, New directional bat algorithm for continuous 685

623 Monarch butterfly optimization based convolutional neural optimization problems, Expert Systems with Appli- 686
co

624 network design, Mathematics 8(6) (2020), 936. cations 69 (2017), 159–175. ISSN 0957-4174. doi: 687

625 [9] E. Tuba, I. Strumberger, T. Bezdan, N. Bacanin and M. https://doi.org/10.1016/j.eswa.2016.10.050. 688

626 Tuba, Classification and feature selection method for med- [22] [22] B. Akay and D. Karaboga, A modified artificial bee 689

627 ical datasets by brain storm optimization algorithm and colony algorithm for real-parameter optimization, Informa- 690

628 support vector machine, Procedia Computer Science 162 tion Sciences 192 (2012), 120–142. ISSN 0020-0255. doi: 691
Un

629 (2019), 307–315, 7th International Conference on Infor- https://doi.org/10.1016/j.ins.2010.07.015. Swarm Intelli- 692

630 mation Technology and Quantitative Management (ITQM gence and Its Applications. 693

631 2019): Information technology and quantitative manage- [23] A.A. Ewees, M.A. Elaziz and E.H. Houssein, 694

632 ment based on Artificial Intelligence. Improved grasshopper optimization algorithm using 695

633 [10] M. Zivkovic, N. Bacanin, E. Tuba, I. Strumberger, T. Bezdan opposition-based learning, Expert Systems with Appli- 696

634 and M. Tuba, Wireless sensor networks life time opti- cations 112 (2018), 156–172. ISSN 0957-4174. doi: 697

635 mization based on the improved firefly algorithm, In 2020 https://doi.org/10.1016/j.eswa.2018.06.023. 698

636 International Wireless Communications and Mobile Com-

637 puting (IWCMC) (2020), 1176–1181.