Page 26 - ITU Journal Future and evolving technologies – Volume 2 (2021), Issue 2
P. 26
ITU Journal on Future and Evolving Technologies, Volume 2 (2021), Issue 2
4.4 Latency minimization through load bal‑ it relies on dynamic creation and withdrawal of replicas
ancing and of loading guided by continuous monitoring of data requests coming
from edge nodes of the underlying network. The pro‑
To reduce high data traf ic in edge networks, Zhao et posed algorithm uses geographicallocation of data during
al. [46] proposed a solution based on their Enumera‑ the distribution process resulting from the plethora of
tion based Optimal Placement Algorithm (EOPA) and ordinary data requests that stem from the clients within
Divide‑and‑Conquer based Near Optimal Placement surroundings. The cost of storing replicas as well as
Algorithm (DCNOPA) to ef iciently distribute virtual expected delay improvement to make a migration or
machine replica copies (VRCs) of applications to the duplication decision to one of the neighboring nodes
edge network. In this solution, a graph ( , ) is is evaluated through the algorithm, which also enables
used to model the physical edge network. The in users to handle the balance between cost minimization
the graph is de ined as { = { , = 1, 2, … , ∣ ∣} and delay optimization. Also, a replica discovery method
and represents the set of edge servers, and the set of where the important nodes are identi ied and noti ied of
connections between edge servers with the assumption replica creations or removals is provided by the proposed
that all mobile users are assigned in the edge network work. The algorithm is complemented with a replica
randomly. Also, the assignment of virtual machines is discovery method where concerned nodes are noti ied of
done according to the following constraints: each virtual nearby replicas. On the other hand, experimental results
machine replica of an application can only be associated show that communication overhead and miscommuni‑
with one edge server, similarly, each edge server only cation errors caused by replica placement and discovery
holds one virtual machine replica of an application. are not signi icant, which is not always true. Also,
The optimal placement algorithm inds the placement the proposed solution is not appropriate for real‑time
′
= { , , ∀ ∈ , ∀ ∈ } among all potential place‑ systems, which require real‑time performance guarantee.
ments of VRCs, to obtain a reduced data traf ic for each
request by considering all potential placement cases for A task scheduling algorithm for MCC based on a heuristic
VRCs, and computing the average data traf ic for each ant colony optimization algorithm was proposed by
placement case. The divide‑and‑conquer based near Wang et al. [49], taking into consideration four types of
optimal placement algorithm divides all edge servers time constrained tasks, adapting to several MCC elements
into clusters and deploys only one VRC for each cluster, such as Cloudlet, mobile device cloud and incorporating
thus reducing the original problem of inding VMs to a variety of objectives including ef icient load balancing,
a problem of determining an ef icient placement for one minimization of energy consumption, and improvement
virtual machine replica in each cluster, which reduces its of reliability and pro it. The proposed algorithm is
complexity considerably. embedded in a system that involves a task tracker, which
is responsible for gathering resource consumption and
Mobile data of loading schemes based on a Finite of loaded tasks information and using the algorithm to
Horizon Markov Decision Process (FHMDP) to reduce determine which task should be executed on a given
the communication cost for delivering mobile data with service provider. It considers four phases or models
different latency sensitivities through several wireless for the resolution of the task scheduling problem. The
networks were proposed by Dongqing and al.[47], where task graph model involves a set of interactive tasks
FHMDP plans data of loading decisions at each decision represented by a graph = ( , ) with representing
epoch. In the model, mobile data is initially delivered to tasks nodes and the relationships between them, with
one or more device through cellular and Wi‑Fi networks. a low that includes tree structure, independent node,
The data being sent from the cloud environment is regular mesh structure and linear chain topology. The
divided into a sequence of data units, which are pre‑ communication model incorporates the channel state
determined by the mobile network operator. Also, the determined by the channel gain and classi ied as good or
access point station that carries a copy of the data can bad depending on a given threshold, the communication
transmit it to the user using D2D communication. The delay de ined as the ratio of the length of each task
approach was embedded in a hybrid of loading algorithm over the channel state. The execution model includes
that can support different delay requirements with lower mobile execution phase that considers the computational
computational complexity. The algorithm computes resource consumption and execution time of each task
the optimal policy through three phases: initialization, de ined by the computing capacity of the device and the
planning and of loading. The expected number of mobile task length, and completion time phase that sums up the
access points in different locations is calculated in the different execution delays of the task. The task scheduling
initialization phase and is used to indicate the availability model considers reduction of resource consumption and
of D2D action in the planning phase, the of loading action pro it maximization for users. The algorithm is divided
at each decision epoch is determined in the last phase. into three parts, known as task selection, which selects
each task to be executed based on the relative pheromone
Aral, Atakan, et al. [48] proposed an algorithm for ratio, service provider selection, which is responsible for
distributed data dissemination and replicas across IaaS; selecting the provider that should execute the selected
12 © International Telecommunication Union, 2021