Page 84 - ITU Journal Future and evolving technologies – Volume 2 (2021), Issue 2

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

4.4.2 Decomposition techniques optimization
methods. Ji and Guo [46] propose a resource

Decomposition techniques are used for challenging prob‑ allocation for two users, one far and one close to the
lems where optimal solutions are likely nonexistent. They MEC server. In the relay mode, where the nearest

decompose the initial problem into sub‑problems, easier user serves as relay between the MEC server and the far
to tackle. They are in addition employed instead of opti‑ user, they employ the Dinkelbach’s method to transform

mal solutions to reduce complexity. A reduced comple- the non‑convex problem into a convex one. Next, they

xity is important in MEC systems where resources are solve it with classical convex optimization methods.
limited unlike in the cloud. They can be a good Wang et al. [55] propose a resource allocation strategy

trade‑off between iciency and results, especially with a two‑stage tandem queues for maximizing the

with systems where medium accuracy is icient. revenue of the network. The irst queue is for packet

Plus, we may implement them easier and in a transmission through the base station and the second

distributed manner. Iterative algorithms When for computational processing at the MEC server. The

initial NP‑Hard problem is decomposed with an Alter‑

considering several joint problems, we can decouple

nating Direction Method of Multipliers (ADMM)‑based

the sub‑problems and solve them individually, like

algorithm into convex sub‑problems. ADMM is an al‑

decoupling the of loading decision and the resource

gorithm to solve problem with a splittable objective
allocation. To retain the connection be‑ tween the

function. It is adapted to decentralized systems because

sub‑problems, we solve them in an iterative algorithm.

of its decomposability and requires a few iterations to

Each iteration takes the output of the previous iteration

converge for modest accuracy [97]. However, it is slow

in input to update the solution until convergence to the

to converge for high accuracy. Yang et al. [60] handle
optimal solution. It allows a decreased complexity but at

the task allocation problem for cloudlets with Bender

the cost of the solution’s precision. In iterative

decomposition to minimize the overall energy con‑

algorithms, we have to pay attention to its conver‑

sumption. Zhang et al. [80] use a ied generalized

gence properties and its required iterations. Li et al. [75] Benders decomposition for latency‑sensitive services

propose a two‑stage heuristic resolving iteratively the of‑ with caching to minimize the overall latency. They also

loading decision and the CPU frequency allocation with solve the problem with a branch and bound method that

the goal of minimizing the energy consumption of mo‑ has an exponential computation complexity. Wang et

bile devices. Pham et al. [49] propose the JOBCA iterative al. [66] introduce MOERA, an online resource allocation
algorithm to solve the resource allocation and of load‑ algorithm to minimize arbitrary operational costs and

ing problem for wireless back‑haul networks. Networks costs that reduce quality of service (e.g., delay) and
with wireless back‑haul may be utilized for rural areas or consider user’s mobility without their prior knowledge.

emergency services where wired back‑haul is expensive They use a regularization technique [98] to divide into

and restrictive. Li et al. [58] introduce an of loading and sub‑problems. Wang et al. [61] consider MEC systems

having a nonblocking state and a blocking state when

resource allocation scheme for multiple wireless access

too much data has accumulated in a server’s buffers.

points to minimize the monetary and energy costs. Tran
For the nonblocking state, they divide the problems into

and Pompili [62] propose a resource allocation scheme

task assignment and resource allocation sub‑problems

for multiservers in ultra‑dense networks to minimize a

with the Cauchy–Schwarz inequality. For the blocking

weighted sum of task completion time with devices’ en‑
state they aim to recover the nonblocking state rapidly

ergy consumption. For that they introduce an iterative

by equalizing the transmitting and computing resource

heuristic algorithm to solve the initial problem in polyno‑

across layers. Lyu et al. [54] address task admission

mial time. Fan and Ansari [65] address the workload al‑

and resource allocation by minimizing the energy con‑

location for cloudlets, considering the trade‑off between

sumption with their EROS scheme. The initial problem
sending tasks to a near cloudlet but overloaded or a far
is simpli ied to an integer programming problem by pre‑
cloudlet but less busy. To simplify the initial problem,

admitting tasks that have to be loaded to meet their

they propose an iterative algorithm solving task assi- deadlines. Then it is resolved with a quanti ied dynamic

gnment and computing resource allocation. Zhang et al. programming algorithm. Haber et al. [77] provide a

[59] aim at inding the trade‑off between latency and resource allocation scheme for UAV‑assisted MEC, taking
energy consumption. They investigate a scenario with into account the UAV positioning and reliability.
one small cell and another with multiple small cells and
4.4.3 Game theory
propose an iterative search algorithm for the multiple
cell scenarios. Zhu et al. [71] introduce a resource
allocation scheme for 5G Industrial Internet of Thing Game theory methods are adapted to systems where each

(IIoT). In this scheme, they include devices with enough node has individuals’ interests. For example, when there

computing resource to help other devices with is a service provider aiming at maximizing its revenue
machine‑to‑machine communication. Mathematical and autonomous devices, each wanting to complete their
decomposition Mathematical solutions exist to tasks as quickly as possible. Theses methods can propose
transform the problem into simpler sub‑problems. a consensus in such systems in a decentralized manner.

70 © International Telecommunication Union, 2021