Page 88 - ITU Journal Future and evolving technologies – Volume 2 (2021), Issue 2

ITU Journal on Future and Evolving Technologies, Volume 2 (2021), Issue 2

resolve these problems by approximating them into con‑ 6.1 Real‑time resource allocation and deploy‑
vex problems iteratively [112]. This method will produce ment
a local optimal solution in a parallel and distributed man‑
MEC application environments evolve and change rapidly,

ner. Some work employs the SCA method to resolve UAV
it is therefore essential to evaluate and predict diverse as‑

trajectory [105, 83, 89, 78] and UAV position [88, 77]

pects of the application and network to respond appro‑

problems. However the resulting optimizer can have a

priately. For the resource allocation scheme, the chan-
high computational complexity and does not respond to
ging traf ic load and channel conditions can hinder the
the real‑time requirement of the system [78].

network, creating bottlenecks and signi icantly impede

the delay. For deployment of mobile resources, user
5.2.3 Greedy algorithms mobility is important to take into account to position the
resource at the proper place and taking account of the
travel time to be sure they are available when needed.
Greedy algorithms are known heuristics solutions for
Many other aspects can impact the resource management
coverage problems [113], such as in UAV deployment.
scheme. Classic machine learning mechanisms may help
They propose a good estimation of the global optimal
predict these aspects, but they have to be meticulously
solution to complex problems. Chen et al. [110] use a
modelled as historic data may not match the application
greedy algorithm to deploy UAVs to locations and
due to the challenging and very fast changing
associate their devices’ tasks to maximize of loaded tasks. environment.
Wang et al. [111] use a greedy algorithm to dispatch

UAVs, considering users’ hotspots, for maximizing the 6.2 Security and privacy

number of processed tasks.
The security and privacy questions in MEC are sensi‑

tive because of the distributed and wireless nature of the
5.2.4 Population‑based meta‑heuristics paradigm. Also, in mission‑critical applications, it is even

more the case as the information can be sensitive and

Population‑based meta‑heuristics search for the best so‑ malicious attackers can take advantage of the situation

lutions in a set of candidate solutions. It starts with a ran‑ or make it worse. The ixed and wearable sensors are
dom population of solutions, then merges, keeps or elimi‑ prone to network attacks on their wireless communica‑

tion. The attackers can jam the communication, rende-
nates each one in each iteration to obtain the most suited.
ring them unreliable or listen to the con idential data.

They have the advantage to avoid local optima [114] at

The cloud is generally more secure than the other

the cost of a higher complexity than a classic optimization
layers of MEC, but privacy is to be considered as we

method. Thus, it can be hard to employ them for online

transmit sensitive data to the Internet. MEC needs

solutions. Besides, each algorithm possesses its own ad‑
vantage and inconvenience. proper security and privacy mechanisms to be reliable in
sensitive situations.
Evolutionary computation Wang et al. [92] use a Diffe-
rential Evolution (DE) algorithm to decide UAV 6.3 Green MEC

location. Their problem possesses a mixed decision
Several pieces of work focus on reducing the energy con‑
variables and is a variable‑length, posing problem to use

sumption of the devices, as it is important to preserve

ef iciently a DE algorithm, so they propose a new their battery. However, they may not consider the

encoding where each UAV in an individual and the
energy consumption on the overall application, i.e., the
population is a deployment solution. Yang et al. [108] energy consumption of the edge and cloud. It is

also use a DE to deploy UAVs at a location to balance the indispensable to consider it globally to achieve green
workload among them to avoid bottleneck in the MEC, therefore minimizing pollution and reduce costs.

network. Further, it is even more the case with mission‑critical

Ions motion optimization Islambouli and Sharafed‑ applications where resources may be on mobile units
and so battery‑constrained.
dine [82] use ions motion optimization [114] to choose

the number of UAVs and their positions, along with de‑

6.4 MEC experimentation and test beds

vice associations and computation allocations. The algo‑

rithm models the population of possible solutions that are The majority of the reviewed work validate their work
anions and cations and choose an ef icient solution itera‑ by simulation. Although there are good simulation tools,
tively. The work [114] shows that ions motions optimiza‑ experiments are valuable to assess a scheme in real si-
tion tend to avoid local optimum and few tuning parame‑ tuations. The prevalence of simulations is
ters, instead of other population‑based algorithms. undoubtedly due to the lack of tools, especially test beds
1

for edge com‑ puting. The SILECS platform proposes a
large‑scale distributed infrastructure from sensors to
6. OPEN ISSUES AND CHALLENGES large data centers, thus making it a possible tool for MEC
experimentation.
In this section, we discuss some still open issues and re‑
lated challenges. 1 https://www.silecs.net/
74 © International Telecommunication Union, 2021