You are on page 1of 7

Available online at www.sciencedirect.

com
Available online at www.sciencedirect.com

ScienceDirect
ScienceDirect
Available online at www.sciencedirect.com
Procedia Computer Science 00 (2021) 000–000
Procedia Computer Science 00 (2021) 000–000
www.elsevier.com/locate/procedia
ScienceDirect www.elsevier.com/locate/procedia

Procedia Computer Science 186 (2021) 235–241

14th International Symposium "Intelligent Systems"


14th International Symposium "Intelligent Systems"
Data
Data processing
processing model
model for
for mobile
mobile IoT
IoT systems
systems
T.T. Aungaa,,*, A.M. Thawaa, N.A. Zhukovab,c, T. Mana, V.V Chernokulskyc
T.T. Aung *, A.M. Thaw , N.A. Zhukovab,c, T. Mana, V.V Chernokulskyc
a
a
ITMO University, Kronverkskiy Prospekt, 49, St Petersburg, 197101, Russia
b
St. ITMO University,
Petersburg Institute forKronverkskiy Prospekt,
Informatics and 49, Stof
Automation Petersburg,
the Russian197101,
Academy Russia
of Sciences,
b
St. Petersburg Institute
14-YaforLiniya
Informatics andStAutomation
B.o., 39, Petersburg,of199178,
the Russian Academy of Sciences,
Russia
c 14-Ya
Saint Petersburg Electrotechnical Liniya B.o.,
University 39, StUlitsa
(LETI), Petersburg, 199178,
Professora Russia
Popova, 5, St Petersburg, 197022, Russia
c
Saint Petersburg Electrotechnical University (LETI), Ulitsa Professora Popova, 5, St Petersburg, 197022, Russia

Abstract
Abstract
Cloud computing has been used for providing services for many years already, but now it is seen that it is not a solution for big
Cloud
moderncomputing
networks has that been usedmultiple
contain for providing services
distributed for many
devices years already,
and produce but nowofitdata.
huge amount is seen
In that
theseit conditions
is not a solution fordata
existing big
modern networks that contain multiple distributed devices and produce huge amount of data. In these
processing models based on cloud technologies suffer from low latency. To overcome this problem, many researchers are conditions existing data
processing
focusing onmodels based onBy
fog computing. cloud technologies
placing fog nodessuffer
in thefrom low latency.
proximity To overcome
of IoT devices, this problem,
many problems havemany researchers
been solved. But are
the
focusing onthe
majority of fogproposed
computing. By placing
fog-based modelsfogare
nodes in the
oriented on proximity of IoTindevices,
data processing many problems
static networks. have be
They cannot been solved. But
effectively usedthe
in
majority of theofproposed
the networks mobile fog-based models
devices, when thearestructure
orientedofonthe
datanetwork
processing in static
changes in networks. They
time. In the cannot
paper we be effectively
propose a newused in
data
the networks of mobile devices, when the
processing model that allow build mobile IoT systems structure of the network changes in time. In the paper we propose a new data
processing
© 2021 Themodel thatPublished
Authors. allow buildbymobile
ELSEVIERIoT systems
B.V.
© 2021
This The
is an Authors.
open accessPublished by ELSEVIER
article under Elsevier B.V.B.V.
the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0)
This is an open access article under the scientific
CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
(https://creativecommons.org/licenses/by-nc-nd/4.0)
Peer-review under responsibility of the committee of the 14th International Symposium “Intelligent Systems”
Peer-review under
Peer-review under responsibility of of the
the scientific
scientific committee of
of the 14th
14th International
International Symposium
Symposium “Intelligent
“Intelligent Systems”.
Keywords: Internet ofresponsibility
Things; data processing; committee
fog computing; dynamic the
networks Systems”
Keywords: Internet of Things; data processing; fog computing; dynamic networks

1. Introduction
1. Introduction
The number of Internet-connected devices is estimated to become over 50 billion by 2020 [1]. These devices,
The number
including of Internet-connected
smartphones, devices
wearable devices, is estimated
sensors to become
and actuators, overcomputers
meters, 50 billioninbyvehicles,
2020 [1]. These
and devices,
so forth, are
including smartphones, wearable devices, sensors and actuators, meters, computers in vehicles, and
connected through networks. They form the Internet of Thing (IoT)-based systems. This attracted many researchers so forth, are
connected through networks. They form the Internet of Thing (IoT)-based systems. This attracted many
to investigate platforms that can manage and analyze the ever-increasing amounts of data produced by the devices. researchers
to investigate
These platformsplatforms that can
can be used manage
as the andbuilding
base for analyzesmart
the ever-increasing amounts
cities, that include manyof datacomponents
smart produced bysuch
the as
devices.
smart
These platforms can be used as the base for building smart cities, that include many smart components
buildings, smart healthcare systems, smart energy, smart mobility, smart transportation, and smart industry. such as smart
buildings, smart healthcare systems, smart energy, smart mobility, smart transportation, and smart industry.

* Corresponding author.
* Corresponding
E-mail address:author.
ttanavy50@gmail.com
E-mail address: ttanavy50@gmail.com
1877-0509 © 2021 The Authors. Published by ELSEVIER B.V.
1877-0509 © 2021
This is an open Thearticle
access Authors. Published
under by ELSEVIER
the CC BY-NC-ND B.V.(https://creativecommons.org/licenses/by-nc-nd/4.0)
license
This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0)
Peer-review under responsibility of the scientific committee of the 14th International Symposium “Intelligent Systems”
Peer-review under responsibility of the scientific committee of the 14th International Symposium “Intelligent Systems”

1877-0509 © 2021 The Authors. Published by Elsevier B.V.


This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Peer-review under responsibility of the scientific committee of the 14th International Symposium “Intelligent Systems”.
10.1016/j.procs.2021.04.143
236 T.T. Aung et al. / Procedia Computer Science 186 (2021) 235–241
2 T.T. Aung et al / Procedia Computer Science 00 (2021) 000–000

Cloud computing is an IT computing infrastructure that has the ability to host and run software services and
applications at a fraction of the cost. The cloud services assume remote decentralized data processing. But
nowadays, cloud computing does not allow meet the requirements imposed to IoT systems. Cloud-based systems
have low latency. So, the researchers considered other scenarios to build latency-sensitive IoT systems. Fog
computing is one of the technologies that can solve latency efficiency problems. Fog computing allow perform real-
time data processing near the data source, that is considered as data processing at the ‘edge’ of the network.
In this paper condition of data processing in IoT systems, background for this work and problem statement are
discussed in section two, general fog architecture and the proposed model are presented in section three, evaluation
is in section four, and the conclusion is in the fifth section.

2. Motivation

2.1. Conditions of Data Processing in IoT Systems

This section defines the role of cloud, fog, middleware computing in data processing in mobile IoT systems.

 Properties of IoT data processing. IoT produces a massive volume of data. It is a heterogeneous data received
from different resources. Usually it is presented in the form of data streams. The data can have different quality.
The main properties of data processing in IoT are the following: processing diverse data structures and data
types, real time data processing, data processing in vehicular and ad hoc networks with dynamic structures. When
processing and storing the data it is necessary to consider spatial and temporal dependencies.
 Challenges in IoT Data Processing. The development of the data processing models for IoT systems is still in the
beginning stage. By using traditional data processing approaches and platforms it is difficult to process IoT data.
The main challenges are: restraints on cooperation of different technologies, limited memory size of IoT devices
that can affect the access to real-time data, context-awareness of processing data provided by different sources,
scalability issues of data integration, detecting fault tolerance in processing systems and support ensuring security
and privacy of user data.

2.2. Background

There are five main architectures of data processing in IoT. They are fog-based data processing, middleware-
based data processing, cloud-based data processing, cloudlets computing, and mobile-edge computing.
Fog based data processing. In fog/edge computing, IoT data are collected from the devices, processed in real-
time, and stored in the cloud. In the work of data stream processing framework (DSPF) for IoT healthcare
applications [2] is proposed. By applying different protocols, data are collected from IoT healthcare devices and
processed in a fog in real time. The benefit of this framework is managing the streaming and processing of the IoT
data. It assumes using fog nodes that are close to data sources. Therefore, latency and energy consumption problems
are solved for limited-resource devices. In the work of the mathematical queuing model for dynamic scaling of fog
nodes [3] is presented. The model can provide answers to the specific IoT questions.
In the work of the car parking system of a city [4] is considered. The system is conducted on the base of Hadoop
MapReduce (MR) platform integrated into fog node. In the proposed architecture there are four layers: device layer
for data acquisition, edge layer for pre-processing, fog layer for data processing and management, and finally, cloud
layer for data interpretation. Data processing is performed on fog nodes.
Middleware-based data processing. Similarly, to fog computing, middleware is located between end devices and
cloud servers to support real-time data processing and analysis. In a distributed middleware architecture [5] is
proposed. In this architecture, there are three layers: data service layer, data filtering and processing layer and the
device layer. The device layer registers all protocols used by different devices and constructs a virtual entity pool for
every device. The entities from the pool are send to the data filtering and processing layer. This layer converts all
heterogeneous data into a uniformed format. Hongyan Cui proposed a blockchain-based decentralized data
management system [6]. According to the experimental results, it effectively defends from unauthorized access and
T.T. Aung
T.T. Aung et al /etProcedia
al. / Procedia Computer
Computer Science
Science 186 (2021)
00 (2021) 235–241
000–000 237
3

empower data provenance and transparency. Latency and energy consumption are affected due to the necessity to
deal with the data management system.
Cloud-based data processing. A stream reasoning system for the smart city applications [7] is proposed. It was
implemented using real-time big data processing technology. In this system Apache Kafka, a message processing
system, Apache Storm, a real-time distributed processing system are used to overcome the constraints associated
with real-time processing. A system called BigPros for medical big data processing [8] was proposed and evaluated.
It is a hybrid cloud system. To generate accurate workflows, the BigPros determines the optimal tradeoff between
the requirement of how many lines of data should be analyzed and the requirement of how fast processing should be
completed. DO Layer performs data management such as data synchronization, ontology generation, data searching,
etc.

Table 1. A summary of the capabilities of existing data processing models in the IoT system.
Energy Data
Models Heterogeneity Scalability Latency
consumption Management
DSPF ✕ ✕ √ ✕ √
Mathematical Queuing Model ✕ ✕ √ √ √
Hadoop MR platform ✕ ✕ √ √ √
Distributed Middleware ✕ ✕ √ ✕ √
Architecture
Decentralized Data ✕ ✕ √ ✕ √
Management Platform
Stream Reasoning System √ √ √ √ √
Model
Workflow Model √ √ √ ✕ √

2.3. Problem statement

Above mentioned research cannot fulfill all the requirements to IoT data processing. The goal of the paper is to
define a model for data processing in IoT networks that allow reaching the required level of the parameters of IoT
systems that are scalability, energy consumption, heterogeneity, and latency. These parameters define the quality of
services provided by IoT mobile systems. We propose a system architecture that provides fog scalability based on
using a queuing model to distribute the load among nodes in the fog layer. By collaboration of fog nodes to process
incoming data, the system can provide the Service level agreement (SLA) that defines the required quality of the
services.

3. Structure of data processing in IoT networks

3.1. General Fog architecture

In this section, we present a three-layer Fog architecture (Fig. 1). Fog architecture is a virtualized architecture
that assumes usage of fog computing between cloud and device layers. It can provide services like a cloud server
because some of the cloud features are running in Fog nodes. It allows overcome latency problems and provide real-
time data processing.
In Fig. 1, FC architecture is composed of three layers: device layer, fog layer, and cloud layer. The lowest layer
or bottom layer is the device layer. IoT devices such as sensors and actuators are placed at this layer. All IoT devices
in this layer continuously generate data that is gathered and send to the system backend for further processing.
The middle layer is the Fog layer. It is composed of the edge computing (EC) devices. These EC devices are
connected to the cloud through the cloud gateway and are responsible for sending data to the cloud domain center
(CDC) periodically. The EC nodes have limited storage capacity that allows them to store the received data
238 T.T. Aung et al. / Procedia Computer Science 186 (2021) 235–241
4 T.T. Aung et al / Procedia Computer Science 00 (2021) 000–000

temporarily. They analyze the data and then send the feedbacks to the source devices. Unlike the traditional cloud
platforms, in FC, all latency-sensitive applications and real-time analysis take place at the fog layer.
The highest layer is the cloud computing layer. The critical components in this layer are the data centers (DCs). A
DC is an infrastructure consisting of physical servers (PSs), storage, and network devices (e.g., switches, routers,
and cables), power distribution systems, and cooling systems. Traditionally DC perform centralized management.
So, it is possible to see the system performance there. DC have rich capabilities for data processing, storage, and
provision of other services.

Fig 1. Three-tier fog architecture

3.2. Data processing model in general Fog architecture

Nowadays many mobile devices are connected to IoT networks. For data processing in mobile IoT networks we
propose to use the model shown in Fig. 2. This model allows build mobile IoT systems that meet the requirements to
energy consumptions, latency, heterogeneity, scalability.

Fig 2. Proposed System Model for Fog Computing


T.T. Aung et al. / Procedia Computer Science 186 (2021) 235–241 239
T.T. Aung et al / Procedia Computer Science 00 (2021) 000–000 5

3.3. Scalability and energy consumption problem solving

At the fog level we propose to apply the queuing model that allow use and manage the queue of incoming data on
the fog nodes. Usage of this model allow solve three problems concerning scalability and energy consumptions. To
solve them we had to find answers on the following questions:

 Given the IoT workload and the available edge computing nodes, what is the QoS offered to users?
 Given the IoT workload and a target QoS, how many edge computing nodes are needed?
 Given the available edge computing nodes and a target QoS, what maximum IoT workload does the system
support?

Fig 3. Fog scalability and load balancing module

When we answered the enumerated questions we assumed that we have X IoT devices. The process of receiving
incoming messages from each of them is a Poisson process with the rate λ. Then the rate of the total process can be
expressed as � � ∑���� λi, where i is an index of the device. As we assume that the processes of receiving incoming
messages from the individual devices are Poisson process, then the aggregate process is also a Poisson process. We
also assume that the incoming messages requests are being serviced using First Come First Served (FCFS) policy.
Let there are N edges computing nodes. The processing times of the EC nodes are independent and identically
distributed. They are exponential random variables with mean rate 1/ EC µ. First, the incoming message which may
carry data or requests is processed by the EC node. After being processed by the EC node, the message is forwarded
to CDC for further processing. Each EC is modeled as an M/M/1/C queue. C is a specified queue length. The
maximum number of messages that each EC can accept and process is C. This implies that a maximum queue length
is C-1. An arriving message joins the queue if it finds less than C request messages in the queue of the EC node
otherwise the message is lost.
The number of fog nodes that we use in our model depends on the workload produced by IoT devices. We
assigned queue length not lower than specific SLA. In our model we use M/M/1/C queue. So, if the messages
coming from devices cannot join the queue of a fog node as its length is more than the maximum queue length, then
fog node will offload the queue, and raw data will be delivered to other nodes by offload method. Because the
number of working fog nodes depends on workload, only the necessary fog nodes will work for data processing.
Consequently, the energy consumption of each fog node and the whole system will decrease significantly.
240 T.T. Aung et al. / Procedia Computer Science 186 (2021) 235–241
6 T.T. Aung et al / Procedia Computer Science 00 (2021) 000–000

3.4. Heterogeneity problem solving

Fog layer has three sub-layers: fog scalability and load balancing layer, middle layer, and data filtering and
processing layer. The middle layer contains an adapter module, protocols module, connection management module,
virtual entity engine, and entity pool [5]. After new messages are received from IoT devices fog scaling and load
balancing are fulfilled. Then the messages are transmitted to the next sub-layer, middle layer. First, they are
processed by protocol registration module. It recognizes the protocol that was used to transfer data stream by the key
information from the stream such as its prex. Then the system queries the entity pool to nd whether the
corresponding entity exists. If not, the virtual entity is created by protocol adapter module and placed into the pool.
Then using the virtual entity the data stream is analyzed. Then the data is sent to the ltering and processing layer,
which is aimed to lter data and convert heterogeneous data to a unied format. When the request on data is
received from the upper layer that is the application layer, it is converted to the corresponding format, which can be
perceived by data consumers. The service layer encapsulates a unied interface that provides access to IoT data or
multiple external systems.

4. Evaluation

We used Java Modelling Tool (JMT) to evaluate the capabilities of the proposed queuing model. The JMT is a
suite of applications for simulating and evaluating the performance of computer systems. JMT contains applications
for systems modelling and for workload analysis. To the first type of applications refer two applications for
simulations that are based on general-purpose queuing models (JSIM). One of them is textual, the other is graphical.
The second type of applications are based on analytical methods. They provide analytical queuing networks with
algorithms (JMVA), Markov chain models underlying queuing systems (JMCH), and bottleneck identification of
queueing network models (JABA). The workload analysis is supported by JWAT.
We evaluated the workload with different numbers of fog nodes 20, 30 and 40, respectively. We assigned arrival
rates for the class equal to 0.01s, service time 0.001s, the maximum number of messages at the queue was 300, and
the maximum number of messages for the whole system was 100000000.

Fig. 4. CPU usage for 20, 30 and 40 fog nodes respectively

Fig. 5. System response time for 20, 30 and 40 fog nodes respectively
T.T. Aung et al. / Procedia Computer Science 186 (2021) 235–241 241
T.T. Aung et al / Procedia Computer Science 00 (2021) 000–000 7

Fig. 6. System throughput for 20, 30 and 40 fog nodes respectively

In Fig. 4 CPU usage for arrival rates is decreasing as the number of working fog nodes is increasing. Figure 5
shows that the system response time for arrival rate changes according to the number of fog nodes. When we use 20
nodes, system response time becomes higher as the arrival rate is increasing. For 30 nodes, system response time is
normal but unstable, and for 40 nodes, response time is more stable. But the throughputs of all number of fog nodes
almost does not change.

5. Conclusion

By using the queueing model at fog level, system allow provide scalability and reduce CPU utility as the
requirements of fog computing nodes depend on the system requests. By placing the entity pool at the fog layer, the
problem of heterogeneity is eliminated.

References

[1] Evans, D. (2011). The internet of things: How the next evolution of the internet is changing everything. CISCO white paper, pp.1-11.
[2] Badidi, E. and Moumane K. (2019). Enhancing the Processing of Healthcare Data Streams using Fog Computing. In: IEEE Symposium on
Computers and Communications. ISCC. Barcelona, Spain, pp.1113-1118.
[3] Kafhali, S.El. and Salah, K. (2017). Efficient and Dynamic Scaling of Fog Nodes for IoT Devices. The Journal of Supercomputing.
[4] Salcic, Z. and Zhang, X. (2018). Big Data Processing in Fog - Smart Parking Case Study. In: IEEE Intl Conf on Parallel & Distributed
Processing with Applications.
[5] Yu, W. and Minbo, L. (2018). An IoT Middleware of Data Service. In: IEEE 11th International Conference on Service-Oriented
Computing and Applications.
[6] Cui, H., Chen, Z., Xi, Y., Chen, H. and Hao, J. (2019). IoT Data Management and Lineage Traceability: A Blockchain-based Solution. In:
IEEE/CIC International Conference on Communications Workshops in China. Changchun, China, pp. 239-244.
[7] Jung, S., Sang, C.Y., Lee, Y.W., Park, J.W. and Yun, C.H. (2017). Cloud Computing Platform Based Real-Time Processing for Stream
Reasoning. In: Sixth International Conference on Future Generation Communication Technologies. FGCT.
[8] Kim, Y.H. and Huh, E.N. (2017). Towards the Design of a System and a Workflow Model for Medical Big Data Processing in the Hybrid
Cloud. In: IEEE 15th Intl Conf on Dependable, Autonomic and Secure Computing, 15th Intl Conf on Pervasive Intelligence and Computing,
3rd Intl Conf on Big Data Intelligence and Computing and Cyber Science and Technology Congress.

You might also like