Page 24 - ITU Journal Future and evolving technologies – Volume 2 (2021), Issue 2
P. 24
ITU Journal on Future and Evolving Technologies, Volume 2 (2021), Issue 2
and potential locations for edge clouds placement and equally developed an algorithm to perform ef icient
represented it by a graph = ∪ , where repre‑ of loading decision in the presence of multiple fog nodes.
sents the base stations, is the set of potential locations Achieving device‑driven intelligence refers to equipping
for edge clouds placement and is the connection devices with smarter functionalities such as sensing,
between two base stations. The steps to solve the edge computing, storage, smart data processing, networking
placement problem are established according to the services and communication; human‑driven intelligence
minimum communication latency between two base associates human domain data with network‑domain
stations and the minimum workload of each edge cloud. decisions that will bene it the network [41].
The scheme takes as input a set of base stations and
edge clouds, and returns the optimal locations of edge The article presents two case studies, user‑behavior‑
clouds. It irst inds out if there is an edge located at a driven healthcare monitoring and device‑driven adaptive
given location, if a base station is allocated to a given task of loading. The irst case study involves using a
edge cloud and if the base station is associated with an machine learning technique‑based health monitoring
edge cloud; then, it de ines a itness function such that module to create a non‑complex ML model that detects
the edge placement problem is transformed into a single human activities driving the sampling of an adaptive sen‑
objective optimization problem by using a weighted sor and scheduling scheme of MAC using some data and
sum method. This problem is solved by selecting the accelerometer sensors. The second case study depicts an
locations with minimal communication delay using K‑ environment involving an end user with N independent
Means algorithm and simplifying the workload allocation tasks, where each task has the possibility to be of loaded
problem using a mixed integer quadratic programming to a computer processor of any of the available fog nodes
algorithm, and then solving it using the Boolean Quadric or processed locally by the end user’s computer proces‑
Polytope cutting plane method. The proposed approach sor; for each task, the user must decide the appropriate
is however not the most ef icient; change of workload CPU to be used to process it with the objective to reduce
size during the allocation is not taken into consideration, delay and energy consumption. The energy consumption
which makes the solution less reliable. and latency minimization problem is a mixed‑integer
nonlinear programming that is solved by irst transform‑
4.3 Energy consumption and latency mini‑ ing the problem into a corresponding uniform Quadratic
mization during data of loading Constrained Quadratic Programming (QCQP), dropping
the rank‑one limitation, which makes the QCQP problem
A system that minimizes execution latency during the SemiDe inite Programming (SDP) convex and can be
migration of a mobile web worker from mobile device to cleared up using the interior point method, and then
an edge server and provides its seamless of loading was constructing a number of reasonable solutions based
proposed by Jeong, Hyuk‑Jin, et al. [40]. In the system, on Gaussian randomization, and inally choosing the
the intact web app that has computation‑intensive codes solution, which minimizes the objective function over all
executed in a web browser, is run by a mobile client. solutions. The shortcoming associated with this solution
When accessible edge servers are detected by the client, is the fact that intelligence in fog computing is still in
the mobile web worker manager is responsible for ind‑ its infancy and the assumptions made are not realistic yet.
ing the best server to process the worker, which reduces
the delay between the time at which a request is sent by Amir Erfan Eshratifar et al. [42] introduced Bottle‑
the main thread to the worker and the time at which a Neck, a new deep learning architecture to reduce the
result is received from the worker. Thus, the HTML5 web workload size to be sent from the UE to the cloud, along
worker is migrated across the cloud, the client, and the with a training method to compensate for the poten‑
edge, and keeps the of loading states while the mobile tial accuracy loss that arises during the compression
client switches its objective server. Web snapshots are of the workload before its transmission to the cloud.
used to move web workers by the system, by a script BottleNeck is basically an auto‑encoder in which the
written in JavaScript to restore the run‑time state of a agent handles the responsibility of learning a compact
web worker when this one is executed. The authors also representation of the features in a transitional layer. It is
highlighted issues of generating a snapshot code that a novel partitioning method that initializes a bottleneck
restores both JavaScript objects and native data such as in a neural network using the suggested BottleNeck unit.
web assembly functions and built‑in objects. Spatial, channel‑wise reduction units and compressor
units are used in its architecture on the mobile device
To reduce energy consumption and latency in fog com‑ to generate a compact representation of the tensor
puting architecture, Quang Duy La et al. [41] proposed that is transmitted to the cloud. BottleNeck’s algorithm
an approach that uses device‑driven and human‑driven comprises three steps, which include training, pro iling
intelligence as key enablers; it performs adaptive low and selection. For a given number of locations in the
latency Medium Access Control (MAC)‑layer scheduling network, BottleNeck is placed on an arbitrary selected
among sensor devices, and detects user behaviors, by layer. Different architectures associated with degrees of
applying machine learning techniques. The authors dimensionality reduction are trained along the channel
10 © International Telecommunication Union, 2021