A Computation Task Offloading Scheme based on

Mobile-Cloud and Edge Computing for WBANs

Rongrong Zhang, Chen Zhou
College of Information Engineering, Capital Normal University, Beijing, China
ICC 2022 - IEEE International Conference on Communications | 978-1-5386-8347-7/22/$31.00 ©2022 IEEE | DOI: 10.1109/ICC45855.2022.9838921

Email: {zhangrr, 2201002079}@cnu.edu.cn

Abstract—The rapid development of Wireless Body Area to provide users with powerful computing resource at a remote
Networks (WBANs) has brought revolutionary changes to the centralized cloud sever [3]. However, the remote cloud server
healthcare system. However, due to the shortcomings of the sink is generally logically and spatially far from users, which
node with limited energy resource and computing capability,
it is difficult to handle all computation tasks effectively and dramatically leads to huge communication and processing
timely. The emergence of Mobile Cloud Computing (MCC) and delay and more energy consumption. Hence, the MCC cannot
Mobile Edge Computing (MEC) may provide a potential and meet the stringent requirement of latency-sensitive applica-
efficient solution. Therefore, we devote this paper to developing tions. As a remedy to these limitations, a new technology,
a computation task offloading scheme based on MCC and MEC called Mobile Edge Computing (MEC), has recently emerged
for WBANs. Technically, we first propose a three-tier system
model with one Remote Cloud Server (RCS), several Mobile Edge to enable in data processing at the network edge, which can
Servers (MESs) and multiple WBAN users. Then, an optimization offer both communication and computational capacities close
problem with the objective to minimize the total cost in terms to users [4]. Due to the distributed and limited computing
of the energy consumption and the delay is formulated. In order capabilities of edge servers, the increasing demand of mobile
to solve the problem, we next investigate a Computation Task applications for high computation and storage resource makes
Offloading Scheme based on Differential Evolution algorithm,
called CTOS-DE. The simulation results demonstrate that our MEC an insufficient approach for accomplishing solely all
proposed CTOS-DE scheme can provide a best computation task the computation tasks. Thus, how to conduct an effective task
offloading decision in terms of the total cost and load balancing. offloading scheme in order to reduce the energy consumption
and satisfy the delay requirement is challenging.
Index Terms—Computation task offloading, Mobile cloud com- In recent years, several research efforts have been dedicated
puting, Mobile edge computing, Wireless body area networks
for the development of computation task offloading. Most of
them focused on offloading tasks on either MCC or MEC.
I. I NTRODUCTION
Specifically, in order to minimize the sum energy consumption
Wireless Body Area Networks (WBANs) with their promis- of users, a resource allocation for a MEC system was studied
ing medical applications in Internet of Things (IoT) have in [5], where the base station made a binary offloading
attracted more and more attention from both academic and decision based on the computing energy and channel gain-
industry recently. Especially in the post-epidemic era, long- s. In [6], the authors attempted to integrate MEC system
term and non-contact monitoring the patients’ physical data with machine learning, where a multi-task learning enabled
through WBANs can dramatically reduce the risk of infection offloading framework for a MEC network was presented to
and workload of medical staff. In a WBAN, a smart phone gen- save the system cost. Subsequently, the authors explored a
erally plays the role of the sink node to collect data from sensor joint optimization of offloading decision and computation
nodes and transmit them to servers via external gateway [1]. resource allocation to minimize the total cost in terms of the
Although the CPU of the smart phones continuously becomes delay and charge in [7]. A game theoretic approach for the
more and more powerful nowadays, many applications running computation offloading decision was proposed in [8], where a
on them still intend to offload computation tasks due to Nash equilibrium could be achieved by designing a distributed
their limited energy resource and computing or hardware computation offloading algorithm.
capabilities [2]. Moreover, the strict delay requirements of Nevertheless, few studies consider the cooperation between
medical applications have become an important obstacle of the MCC and MEC, which can not only balance the load of servers
evolutionary development of WBANs. As a result, the smart but also guarantee the QoS requirements of different users.
phones cannot handle large amounts of computation tasks in In [9], the authors both considered the single and multiple
a short time. In addition, we cannot only rely on the endless users offloading problem, where both the resource constraint
code optimization with the overflow of the massive amounts and the interference among multiple users were taken into con-
of data. Consequently, computation task offloading is one of sideration. An online peer offloading framework by leveraging
the most efficient solutions to relief the energy consumption the Lyapunov technique was developed in [10] to keep the
and execution delay of the smart phone in a WBAN. energy consumption of base stations below a threshold. As hu-
Fortunately, the Mobile Cloud Computing (MCC) has been man mobility can significantly impact the offloading decision,
considered as a potential approach and an effective technique the authors in [11] proposed a mobility-aware cooperative task

Authorized licensed use limited to: SRM University Amaravathi. Downloaded on September 06,2023 at 10:55:11 UTC from IEEE Xplore. Restrictions apply.
4504
offloading scheme to reduce the service time, the percentage
of failed tasks and the energy consumption of WBAN users.
Then, a joint task assignment and power allocation problem
was investigated in [12], where a dynamic offloading decision
was proposed to minimize the total execution latency.
However, most of the existing work mainly considered the
system model with a MCC or/and a MEC with multiple users.
Indeed, there are generally multiple MECs around WBAN
users. Moreover, the QoS requirements of different types of
computation tasks are quite different, which is apparently
overlooked. Motivated by the above observations, we devote
this paper to developing a computation task offloading scheme
to reduce the energy consumption of sink node and guarantee
the delay requirements of medical tasks simultaneously. The
main contributions of this paper are articulated as follows:
• We first develop a three-tier system model with a Remote Fig. 1: The system model
Cloud Server (RCS), several Mobile Edge Severs (MESs)
and multiple WBAN users, where all MESs can connect
with each other to offload computation tasks among them. Mj ∈ MS and Mj 6= Mi . In this paper, we consider that the
• Second, an optimal problem with the objective to mini-
computation task of sink node can be executed locally on itself
mize the total cost in terms of the energy consumption CPU or be fully offloaded to process on the RCS or one MES.
and delay is formulated, where the queuing delay is taken Let ri ∈ {0, 1, ..., M, M + 1} represent the computation task
into consideration. offloading decision of Si , thus we can have
• Third, we propose a Computation Task Offloading 
 0 Local execution
Scheme based on Differential Evolution algorithm 

M
(CTOS-DE) to provide an optimal offloading decision. i Local MES execution
ri = . (1)
• Finally, the simulations are further conducted to demon- 

 Mj Other MES execution
M +1

strate that our CTOS-DE scheme performs better in terms Remote RCS execution
of the total cost and load balancing among all MESs.
When ri = 0, it means that the task of Si will be executed on
II. S YSTEM M ODEL Si locally. As ri = Mi , it expresses that the task of Si will be
In this section, we first describe the system model. And the offloaded to the local MES Mi . Correspondingly, the task of
channel model is then introduced. Si will be processed on the MES Mj , if ri = Mj . Otherwise,
the task will be uploaded to the RCS when ri = M + 1.
A. System Model
In this paper, we consider a three-tier system model consist- B. Channel Model
ing of a Remote Cloud Server (RCS), M Mobile Edge Servers For medical applications in IoT, WBANs can monitor and
(MESs) and N WBAN users, as shown in Fig.1. Assume transmit the physiological parameters in complex environment
that the RCS can provide virtually unlimited available energy such as a hospital building. In this paper, we consider in
and computing resource. The MESs yet with sufficient energy a hospital environment where patients are concentrated, and
resource and limited computational resource can communicate there are many kinds and numbers of obstacles which will
with the RCS directly and connect with each other through greatly affect the wireless link quality. Thus, the pass loss
optical fiber link. As there is one and only one sink node in a between the sink node Si and its local MES Mi can be
WBAN, and we consider the full task offloading of a WBAN, calculated as
it is interchangeable between a ”sink node” and a ”WBAN”
in this paper. Without loss of generally, the sink node which is
generally played by a smart phone with insufficient energy and dSi Mi X
L(dSi Mi ) = L0 (d0 ) + 10nlg( )+ Kg Lg ,
computing resource can communicate with its local MES by d0 g
wireless link. Notice that a MES can cover multiple WBANs,
a WBAN belongs to only one local MES, and a WBAN cannot where dSi Mi is the distance between Si and MES Mi . L0 (d0 )
communicate with the RCS directly. denotes the pass loss when the reference distance is d0 , i.e.,
Let MS = {1, 2, ..., M } index the set of all MESs and L0 (d0 ) = 10lg(4πd0 /λ)2 , λ is the wireless signal wavelength.
SS = {1, 2, ..., N } denote the set of all sink nodes. For each n represents the path loss factor. Kg and Lg define the number
sink node Si (i ∈ SS), we define Mi is the ID of its local and path loss of the g-th obstacle, respectively.
MES, which means that Si can communicate with Mi directly. Given the transmission power PiT of sink node Si , the pass
Correspondingly, let Mj express one of other MESs, that is loss L(dSi Mi ) can also be given by

Authorized licensed use limited to: SRM University Amaravathi. Downloaded on September 06,2023 at 10:55:11 UTC from IEEE Xplore. Restrictions apply.
4505
 MES
where fM is the computation capability of the MES Mi .
PiT i
And Cqk (qk ∈ SS) is the total number of CPU cycles of the
L(dSi Mi ) = 10lg R ,
PMi k-th computation task in the task queue of the MES Mi .
R Next, the execution time of computation task τi on the MES
where PM is the received power of the MES Mi .
i Mi is given by
According to the Shannon equation [13], the transmission Ci
rate VSi Mi between Si and MES Mi can be calculated as TiEMi = MES . (6)
fMi
R
PM i
Therefore, the processing time of computation task τi exe-
VSi Mi = BSi Mi log2 (1 + ), (2)
N0 cuted on the the MES Mi can be formulated as
where BSi Mi is the wireless channel bandwidth, and N0 TiMi = TiSi Mi + TiW Mi + TiEMi . (7)
denotes the background noise power.
3) Other MES execution: When the offloading decision
III. P ROBLEM F ORMULATION ri = Mj , the computation task τi will be offloaded on the
In this section, we first introduce the processing time CPU of the MES Mj . In this situation, the processing time of
model and energy consumption model. Then, the problem computation task τi includes four parts: the transmission time
formulation is described. from Si to MES Mi , the transmission time from MES Mi to
MES Mj , the waiting time of the task queue of the MES Mj
A. Processing Time Model and the execution time on the MES Mj .
We consider that sink node Si has a computation task τi = Inspired by the above analysis of cases 1) and 2), we can
(Pi , Ci , TiLimit ) that can be executed either locally on the sink accordingly have
node or remotely on the RCS or one MES via computation M Pi WM EM
offloading. Here Pi denotes the size of computation task τi , Ti j = TiSi Mi + + Ti j + Ti j , (8)
VMi Mj
and Ci defines the total number of CPU cycles required to
where VMi Mj represents the transmission rate between the
accomplish the computation task τi . As the computation tasks
MES Mi and the MES Mj .
can be simply classified into urgent or non-urgent in WBANs,
4) Remote RCS execution: As the offloading decision ri =
let TiLimit denote the maximum tolerance execution time of
M + 1, the computation task τi will be offloaded to the RCS.
computation task τi . According to analysis in Eq.(1), we then
Correspondingly, the processing time of computation task τi
deliberately discuss the processing time model of computation
is made up of the transmission time from Si to MES Mi and
task τi in four cases.
from MES Mi to the RCS and the execution time on the RCS,
1) Local execution: When the offloading decision ri = 0,
which can be elaborately expressed as
the computation task τi will be locally executed on the CPU
of sink node Si . Let fisink be the computation capability (i,e., Pi Ci
TiRCS = TiSi Mi + + RCS . (9)
CPU cycles per second) of sink node Si . Note that different VMi RCS f
sink nodes may have different computation capabilities. Thus, VMi RCS is the transmission rate between the MES Mi and the
the processing time of computation task τi by local execution RCS, and f RCS denotes the computation capability of the RCS.
is given as In summary, by merging the four cases, the processing time
Ci of computation task τi can be formulated as
Tisink = sink . (3)
fi Mj
Ti = αTisink + βTiMi + µTi + γTiRCS , (10)
2) Local MES execution: When the offloading decision
ri = Mi , the computation task τi will be offloaded on the where α + β + µ + γ = 1 and α, β, µ, γ ∈ {0, 1}. Notice that
CPU of local MES Mi who can communicate directly with the processing time Ti cannot be larger than the maximum
sink node Si . In this case, the computation task τi needs firstly tolerance latency TiLimit .
to forward to the local MES Mi . Given the transmission rate B. Energy Consumption Model
VSi Mi in Eq.(2), the transmission time of computation task τi
As the energy of sink node is limited, we focus on analyzing
can be calculated as
the energy consumption model of sink node for each offloading
Pi decision. For case 1) that the computation task τi is locally
TiSi Mi = . (4)
VSi Mi executed on the sink node Si , the energy consumption can be
Then, we derive the waiting time of computation task τi presented as
in the task queue of the MES Mi , which depends on the Eisink = Tisink · PiE , (11)
offloading decision of all sink nodes. Assume that computation where PiE defines the execution power of sink node Si .
task τi is the b-th task processed by MES Mi . Based on the Correspondingly, as the computation task τi is offloaded on
Queuing Theory, the waiting time can be computed by the local MES Mi in case 2), given the waiting power PiW of
Pb−1
Cqk sink node Si , we can derive the energy consumption as
W Mi
Ti = k=1 , (5)
MES
fM i
EiMi = TiSi Mi · PiT + (TiW Mi + TiEMi ) · PiW . (12)

Authorized licensed use limited to: SRM University Amaravathi. Downloaded on September 06,2023 at 10:55:11 UTC from IEEE Xplore. Restrictions apply.
4506
 Subsequently, for case 3) when the computation task τi is Algorithm 1 Computation task offloading scheme (CTOS-DE)
forwarded on the MES Mj , the energy consumption can be Input: Lmax , τi , fisink , fm
MES
, f RCS for i ∈ SS, m ∈ MS
calculated as Output: Υbest
M Pi 1: L = 1
Ei j = TiSi Mi · PiT + ( + TiW Mi + TiEMi ) · PiW . (13)
VMi Mj 2: Randomly generate T group offloading decisions

If the computation task τi is executed on the RCS, we thus RT = [Υ1 , . . . , Υt , . . . , ΥT ] where Υt = [r1 , . . . , rN ]
3: for all m ∈ MS do
have
4: for each t → T do
Pi Ci
EiRCS = TiSi Mi · PiT + ( + RCS ) · PiW . (14) 5: Record the task queue Qmt
VMi RCS f 6: end for
As a consequence, the energy consumption model of the 7: end for
sink node Si can be finally given by 8: for each t → T do
Mj 9: for all i ∈ SS do
Ei = αEisink + βEiMi + µEi + γEiRCS . (15)
10: Compute time Tit and energy consumption Eit
C. Problem Formulation 11: if Tit > TiLimit then
0

In medical applications, different tasks have the different 12: RT =RT -Υt
delay and energy consumption requirements. Generally, for the 13: end if
urgent tasks, we must preferentially ensure the processing time 14: end for
15: end for
requirement. Nevertheless, the energy consumption is prior to 0
16: for each t → T do
the delay for the non-urgent tasks. In order to simultaneously
satisfy the delay and energy consumption requirements, we 17: Compute the total cost St
18: end for
define the weighted total cost of the computation task τi as 0 0
19: Υbest =min S in RT
Ti Ei rt
Si = ωi sink
+ ϕi sink , (16) 20: if L > Lmax then
Ti Ei 0
21: Υbest = Υbest and return Υbest
where ωi , ϕi ∈ [0, 1] and ωi + ϕi = 1 for ∀i ∈ SS. ωi and ϕi 22: else
are the weight factors of the delay and the energy consumption 23: for each t → T do
to sink node Si . 24: Select two offloading decisions Υt1 and Υt2 in RT
0 0
Without loss of generally, the computation task offloading 25: Υt = Υbest +F (Υt1 − Υt2 )
for multiple WBANs based on mobile cloud and edge com- 26: end for
puting can be formulated as an optimization problem. The 27: for each t → T do
objective is to minimize the sum weighted total cost of all 28: for all i ∈ SS do
00 0
sink nodes under the constraints of the processing latency 29: rt = rand (rt , rt )
and the computation capacities of the RCS, MES and sink 30: end for
node servers. Mathematically, the optimization problem can 31: end for
be expressed as 32: for each t → T do
N 33: Υt = min00 S
X rt ,rt
min Si 34: end for
ri
i=1
35: end if
s.t. ri ∈ {0, 1, · · · , M, M + 1} . (17) 36: L = L + 1
Ti < TiLimit , i ∈ SS 37: Repeat 3-36
fisink < fm
MES
< f RCS , m ∈ MS
IV. C OMPUTATION TASK O FFLOADING S CHEME BASED ON
D IFFERENTIAL E VOLUTION A LGORITHM Task Offloading Scheme based on the Differential Evolution
In this section, we propose a computation task offloading algorithm (CTOS-DE) to solve the problem (17).
scheme based on the Differential Evolution (DE) algorithm to More specifically, each sink node computes the transmission
solve the optimization problem (17). rate between itself and its local MES according to the channel
As the offloading decisions of every sink node in our system model in section II-B, and then forwards the task information
model are interacted on each other, it turns out unfortunate- including the computation task size, the total required CPU
ly that the optimal problem (17) is NP-hard [8] which is cycles number and the maximum tolerance latency to the RCS
extremely challenging to solve mathematically. However, the through its local MES, in the first phase. In the next phase, the
DE algorithm [14] has been proved that it is an efficient and RCS executes our CTOS-DE scheme which will be formally
powerful stochastic search technique for solving optimization described in Algorithm 1 after receiving all the computation
problems. Therefore, we develop a three-phases Computation tasks, and broadcasts the final offloading decisions to all sink

Authorized licensed use limited to: SRM University Amaravathi. Downloaded on September 06,2023 at 10:55:11 UTC from IEEE Xplore. Restrictions apply.
4507
nodes. Subsequently, each sink node offloads its computation Start

task to the corresponding server according to the final offload-

Input: Lmax , τi , fisink , fm
MES
, f RCS
ing decisions during the third phase.
A formal description of the CTOS-DE is shown in the
Generate offloading decisions RT
Algorithm 1, and we also offer an operation flowchart in
Fig. 2 for the sake of clarity. Specifically, we initialize the
Record the task queue Qmt
system parameters and the task information for all sink nodes,
all MESs and the RCS. Then, the RCS randomly generates Compute time Tit and energy consumption Eit
T group offloading decisions RT for all sink nodes, and it
records the task queue information Qmt for all MESs. Under Compute the total cost St and select Υbest
0

each offloading decision, the RCS respectively computes the

processing time Tit and the energy consumption Eit based on
Eq.(10) and Eq.(15) for all sink nodes as lines 8-15 in the Yes 0
L > Lmax Υbest = Υbest
Algorithm 1. If the processing time Tit exceeds the maximum
latency TiLimit , we need deleting the t-th offloading decision No Output: Υbest
Υt from RT . Next, we compare the total cost St among Mutation operation: get
0
RT as lines 23-26
0
all offloading decisions in RT , and select a tentative best End
0
00
offloading decision Υbest with the minimum total cost as in Crossover operation: get RT as lines 27-31
0
lines 16-19. Subsequently, let Υbest = Υbest and return the final
best offloading decision Υbest if the iteration number satisfies Selection operation: update RT as lines 32-34
the stopping criterion. Otherwise, we randomly choose two
offloading decisions Υt1 and Υt2 in RT to accomplish the L = L+1
0 0
mutation operation, i.e.,Υt = Υbest +F (Υt1 − Υt2 ) where F
Fig. 2: The flowchart of the CTOS-DE scheme
is a scaling factor, as in lines 23-26. As a result, we can
0
obtain another T group offloading decisions RT . After that,
it achieves the crossover operation to renew the offloading has its own MES with 20 GHz CPU frequency. There is only
0
decision by randomly selecting a value between rt and rt for one RCS who is far from WBANs and whose computation
each sink node under each offloading decision in lines 27- capability is 50 GHz. Assume that there are many types of
31. Consequently, we can have the third T group offloading obstacles in a hospital, such as concrete wall, brick wall and
00
decisions RT . In order to make the set of offloading decisions glass wall with pass loss 10.8, 2.5∼6.0 and 2.31, respectively.
even better, we finish the selection operation and update the In addition, the wireless signal frequency is 2.4 GHz, and the
Υt by choosing the one that can minimize the total cost S in rt noise power is set to -100 dBm. Moreover, the other simulation
00
and rt as shown in lines 32-34. Thus, we gain a fresh T group parameters are given in Table I. Furthermore, every simulation
offloading decisions RT which absolutely perform better in result is the average of 100 independent experiments.
terms of the total cost than the initialized offloading decisions.
In the wake of the fresh RT , we repeat the operations as lines TABLE I: Simulation parameters
3-36 until it meets the stopping criterion, and then output the
final offloading decision Υbest . Parameter Value
Here, we discuss the complexity of Algorithm 1. In each Computation task size 300 ∼ 1500 KB
loop, for given T group offloading decisions, we need to record Required CPU cycles (1 ∼ 3)×108 cycles
the task queue of each MES, which results in the complexity Maximum tolerance latency 0.1 ∼ 3 s
as O(T M ). Then, the mutation, crossover and selection oper- CPU frequency of sink node fisink 0.3 GHz
MES
CPU frequency of MES fm 20 GHz
ations are executed among all sink nodes, thus the correspond-
ing complexity can be given by O(T ) + 2 · O(T N ). Given the CPU frequency of RCS f RCS 50 GHz
maximum loop number Lmax , the complexity of Algorithm 1 Waiting power of sink node PiW 100 mW
can be approximated by O(Lmax T M ) + O(Lmax T N ). Transmission power of sink node PiT 200 mW
Execution power of sink node PiE 500 mW
V. P ERFORMANCE E VALUATION
In this section, we evaluate the performance of our proposed We compare the proposed CTOS-DE scheme with two
CTOS-DE scheme in terms of the total cost and load balancing reference algorithms: (1+1) Evolution Strategy ((1+1)ES) [15]
among all MESs. In the following, we first introduce the and Genetic Algorithm (GA) [12] which are main computation
simulation settings, and then present the simulation results. offloading algorithms based on the cooperation between MCC
and MEC. Specifically, (1+1)ES is an evolution strategy where
A. Simulation Settings the offspring individuals evolve from one parent. GA is a
In our simulations, we consider a scenario where the WBAN meta-heuristic algorithm which optimizes the population by
users are randomly located in the hospitals and each hospital iterations.

Authorized licensed use limited to: SRM University Amaravathi. Downloaded on September 06,2023 at 10:55:11 UTC from IEEE Xplore. Restrictions apply.
4508
 Fig. 3: The total cost S Fig. 4: The average length of task queue

B. Simulations Results ACKNOWLEDGMENT

This work is supported by the National Natural Science
We first assess the total cost with the increasing of the
Foundation of China (No.62171296) and the Science and
number of MESs when the urgent tasks occupy half of all tasks
Technology Project of Beijing Municipal Education Commis-
as shown in Fig.3. In detail, the weight factor of the delay ω
sion (No.KM202010028005).
is set to 0.8 and that of the energy consumption ϕ is 0.2 for
urgent tasks. The corresponding factors for non-urgent task R EFERENCES
are 0.2 and 0.8, respectively. Obviously, Fig.3 shows that our [1] S. Sodagari, B. Bozorgchami, and H. Aghvami, “Technologies and
proposed CTOS-DE scheme performs best in terms of the total challenges for cognitive radio enabled medical wireless body area
cost among the three algorithms. Moreover, the superiority networks,” IEEE Access, vol. 6, pp. 29567–29586, 2018.
[2] D. Sabella and A. Vaillant, “Mobile-edge computing architecture: The
of our scheme is outstanding when the number of MES is role of mec in the internet of things,” IEEE Consumer Electronics
small. Specifically, the total cost of the CTOS-DE is reduced Magazine, vol. 5, no. 4, pp. 84–91, 2016.
by 11.2% and 13.9% than that of GA and (1+1) ES when the [3] Z. Zhang and S. Li, “A survey of computational offloading in mobile
cloud computing,” in 2016 4th IEEE International Conference on Mobile
number of MES is equal to 10, respectively. Cloud Computing, Services, and Engineering, pp. 81–82, 2016.
Fig.4 illustrates the average length of task queue of MES [4] N. Abbas, Y. Zhang, A. Taherkordi, and T. Skeie, “Mobile edge
computing: A survey,” IEEE Internet of Things Journal, vol. 5, no. 1,
against the increasing of the number of WBANs. It is shown pp. 450–465, 2018.
that our CTOS-DE scheme can achieve the shortest average [5] C. You, K. Huang, H. Chae, and B.-H. Kim, “Energy-efficient resource
length of task queue among the three algorithms. A quantita- allocation for mobile-edge computation offloading,” IEEE Transactions
on Wireless Communications, vol. 16, no. 3, pp. 1397–1411, 2017.
tive comparison can be made that the length of task queue [6] B. Yang, X. Cao, J. Bassey, X. Li, T. Kroecker, and L. Qian, “Computa-
is only increased by 58.8% when the number of WBANs tion offloading in multi-access edge computing networks: A multi-task
increases from 300 to 600. However, the values of (1+1) ES learning approach,” in IEEE International Conference on Communica-
tions (ICC), pp. 1–6, 2019.
and GA schemes enlarge to 83.7% and 122.7% respectively. It [7] K. Wang, Z. Hu, and Q. Ai, “Joint offloading and charge cost min-
further demonstrates the superiority of our CTOS-DE scheme imization in mobile edge computing,” IEEE Open Journal of the
in terms of the load balancing among all MESs. Communications Society, vol. 1, pp. 205–216, 2020.
[8] X. Chen, L. Jiao, W. Li, and X. Fu, “Efficient multi-user computation
According to the above analysis, we can draw a conclusion offloading for mobile-edge cloud computing,” IEEE/ACM Transactions
that our CTOS-DE scheme can provide a better computation on Networking, vol. 24, no. 5, pp. 2795–2808, 2016.
task offloading decision than that of (1+1) ES and GA algo- [9] Z. Ning, P. Dong, and X. Kong, “A cooperative partial computation
offloading scheme for mobile edge computing enabled IoT,” IEEE
rithms, in terms of the total cost and load balancing among Internet of Things Journal, vol. 6, no. 3, pp. 4804–4814, 2019.
all MESs. [10] L. Chen, S. Zhou, and J. Xu, “Computation peer offloading for energy-
constrained mobile edge computing in small-cell networks,” IEEE/ACM
VI. C ONCLUSION Transactions on Networking, vol. 26, pp. 1619–1632, 2018.
[11] Y. Liao, Y. Han, Q. Yu, Q. Ai, Q. Liu, and M. S. Leeson, “Wireless body
In this paper, we have studied a computation task of- area network mobility-aware task offloading scheme,” IEEE Access,
vol. 6, pp. 61366–61376, 2018.
floading scheme for WBANs. First, we introduced a three- [12] U. Saleem, Y. Liu, and S. Jangsher, “Mobility-aware joint task schedul-
tier framework with one RCS, several MESs and multiple ing and resource allocation for cooperative mobile edge computing,”
WBANs. Then, an optimization problem with the objective to IEEE Transactions on Wireless Communications, vol. 20, no. 1, pp. 360–
374, 2021.
minimize the total cost formulated. Subsequently, we proposed [13] C. E. Shannon, “A mathematical theory of communication,” The Bell
a computation task offloading scheme based on the differential System Technical Journal, vol. 27, no. 3, pp. 379–423, 1948.
evolution, denoted by CTOS-DE, to solve the problem. Finally, [14] R. Storn, “Differential evolution-a simple and efficient heuristic for glob-
al optimization over continuous space,” Journal of Global Optimization,
the simulation results demonstrated that our CTOS-DE scheme vol. 11, 1997.
performed better than that of (1+1) ES and GA algorithms in [15] T. Glasmachers, “Global convergence of the (1+1) evolution strategy to
terms of the total cost and load balancing. a critical point,” Evol. Comput., vol. 28, p. 27C53, Mar. 2020.

Authorized licensed use limited to: SRM University Amaravathi. Downloaded on September 06,2023 at 10:55:11 UTC from IEEE Xplore. Restrictions apply.
4509