Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
ISSN 2383-630X(Print) / ISSN 2383-6296(Online)
Journal of KIISE, Vol. 45, No. 4, pp. 395-402, 2018. 4
https://doi.org/10.5626/JOK.2018.45.4.395
․This work was supported by Institute for Information & communications
Technology Promotion(IITP) grant funded by the Korea government
(MSIT) (No.2015-0-00557, Resilient/Fault-Tolerant Autonomic Networking
Based on Physicality, Relationship and Service Semantic of IoT Devices)
․이 논문은 2017 한국컴퓨터종합학술 회에서 ‘Smart Agent Based Data
Aggregation for Smart City’의 제목으로 발표된 논문을 확장한 것임
논문 수 : 2017년 7월 31일
(Received 31 July 2017)
심사완료 : 2018년 1월 25일
(Accepted 25 January 2018)
CopyrightⒸ2018 한국정보과학회ː개인 목 이나 교육 목 인 경우, 이 작물
의 체 는 일부에 한 복사본 혹은 디지털 사본의 제작을 허가합니다. 이 때,
사본은 상업 수단으로 사용할 수 없으며 첫 페이지에 본 문구와 출처를 반드시
명시해야 합니다. 이 외의 목 으로 복제, 배포, 출 , 송 등 모든 유형의 사용행
를 하는 경우에 하여는 사 에 허가를 얻고 비용을 지불해야 합니다.
정보과학회논문지 제45권 제4호(2018. 4)
†
††
†††
††††
학생회원
비 회 원
학생회원
종신회원
:
:
:
:
경희 학교 컴퓨터공학과
경희 학교 컴퓨터공학과
경희 학교 컴퓨터공학과
경희 학교 컴퓨터공학과 교수(Kyung Hee Univ.)
(Corresponding author임)
지연에 민감한 스마트 시티 서비스를 한 스마트 에이 트 기반의
동 데이터 수집 기법 연구(Smart Agent based Dynamic Data Aggregation for
Delay Sensitive Smart City Services)
엠디 시라 무니르† 살더 크룰 아베딘
†† 엠디 골람 라비울 알람
††
(Md. Shirajum Munir) (Sarder Fakhrul Abedin) (Md. Golam Rabiul Alam)
김 도†††
홍 충 선††††
(Do Hyeon Kim) (Choong Seon Hong)
요 약 스마트시티는 녹색기술과 더불어 지속가능한 사회 발 을 한 의 지능형 기술의 비 이
다. IoT 기반의 스마트서비스, 를 들어 스마트 운송, 스마트 건강, 스마트 홈, 스마트그리드, 스마트 보안
등의 응용 로그램은 질높은 삶과 웰빙생활을 보장하는 스마트시티의 핵심 원동력이다. 이러한 서비스들을
가능하게 하기 해, 애 리 이션은 다수의 IoT노드로부터 데이터를 수집해야한다. 이 경우, 스마트 시티
의 앙 집 식 네트워크에서 거 한 네트워크 트래픽을 리하는 것이 더 어려워진다. 본 논문에서는 스
마트시티의 분산된 고 도 네트워크에서 Dynamic Data Aggregation기반의 스마트에이 트를 도입하여
여러 서비스 제공 업체 명한 시민들의 서비스 요청을 충족시킴으로써 이 문제를 해결하 다. 마지막
으로 제안하는 스마트 에이 트 기반 동 데이터 집 모델의 결과를 시뮬 이션 한 결과, 서비스 이행
시간과 컨버 스와 련하여 제안된 근법에 한 성능이 향상되었다.
키워드: 스마트시티, 동 데이터 수집, 스마트 에이 트, 분산 토폴로지, 사물인터넷
Abstract Smart city is the vision of modern intelligent technology toward the sustainable
development of green technology and social development. Smart services e.g. smart transportation,
smart health, smart home, smart grid, smart security, and IoT based applications are the key enablers
of smart city, that ensure the quality life and well-being. In a bid to ensure the functionalities of those
services, the IoT applications gather data from numerous IoT nodes. In such a case, it becomes more
396 정보과학회논문지 제45권 제4호(2018. 4)
challenging to managing huge network traffic in the centralized network of smart city. Therefore, in this
research, we have focused on the resolution of this problem through the introduction of of smart
agent-based dynamic data aggregation (DDA) from distributed dense smart city network for city service
fulfillment. In this research study, we purposed to model a peer to peer fully distributed system using
distributed hash table chord protocol. We also proposed an algorithm for the IoT network and designed
smart agent based IoT node searching algorithm for crowd sourcing. Finally, we simulated the result of
the proposed smart agent based dynamic data aggregation model in an effort to achieve a higher
performance gain for the proposed approach in respect to service fulfillment time and convergence.
Keywords: smart city, dynamic data aggregation, smart agent, distributed topology, internet of things
1. Introduction
Smart city shall be able to integrate understandably
and seamlessly an enormous number of heterogeneous
sensors and actuators with microcontroller based IoT
devices. The existing research on IoT based smart
city includes defining IoT architectures, protocols
along with a wide variety of enabling technologies
[1-3]. The data center in the cloud is linked to a set of
services, such as electrical energy, water, and gas supply
provided in smart cities. Additionally, the smart agents
are goal oriented, autonomous, adaptive, and coopera-
tive to achieve the application specified targets [4].
Smart city like environment is comprised of more
than a billion of IoT nodes. In Figure 1, we consider
a typical smart city geographic area with densely
scattered access points (APs) (red diamond) and IoT
nodes (blue dot) where 51 access points (AP) and
2601 IoT nodes. It becomes challenging to provide
quick response for a single point controlled network
when the request are a massive amount. In a generic
system model of a smart city, first, all the IoT nodes
transmit data to the corresponding cluster head in a
certain geographic area. Then, the cluster head takes
a decision for sending the IoT sensor data to a
single point receiver based on the data priority [5].
However, the major problem of a centralized network
is, if the network somehow collapses, the whole
smart city network clamps down.
Fig. 1 APs (Diamond) and IoT Nodes (Dot)
For the smart city security system, there are many
surveillance cameras in the city which are connected
with some APs. In a typical IoT network, if the
security manager needs to monitor the real-time data
feed of 10 surveillance cameras, the manager needs
to send a request to the server or central AP. As a
result, there is an additional traveling cost for the
APs with a substantial amount of queuing delay.
Furthermore, the IoT network also depends on a
single point AP. Now the research question is if the
centralized AP is collapsed due to a large volume of
data request, how the AP will provide services for
the requester.
In this paper, we have proposed a fully distributed
peer to peer chord network for the smart city and a
smart agent-based dynamic data aggregation technique
for the smart city services. When there is a service
fulfillment or data collection request, the smart agent
is responsible for finding the AP list for the corre-
sponding IoT nodes from the chord lookup table and
it calculates the minimum traverse cost between the
APs with the individual delay cost. Finally, the smart
agent travels through the minimum cost APs for the
request fulfillment. This research work is extended
from our previous paper [6].
The rest of the paper is organized as follows:
Section 2 presents the background and current
research of smart city services whereas Section 3
provides the system model of the proposed model.
In Section 4 depicts the problem formulation and
algorithm design. Simulation and performance ana-
lysis are included in Section 5. Finally, we provide
the concluding remarks in Section 6.
2. Background
Mobility, transportation, public services, supply
chain management, security management are the
지연에 민감한 스마트 시티 서비스를 한 스마트 에이 트 기반의 동 데이터 수집 기법 연구 397
Fig. 2 System Model
basic components of a smart city like urban environ-
ment. In the smart cities, it is absolutely essential to
managing the natural resources for smart recycling
and usage. By the increasing number of smart city
application, the volume of multi-dimension IoT
sensor data needs to be processed in a very short
time period for ensuring better Quality of Service
(QoS) [7]. Previously, most of the smart city networks
are being managed in a centralized manner or based
on cluster head.
The traditional service fulfillment approach is to
find all-pairs shortest paths and construct a simp-
lified graph for the candidate nodes. By using a new
graph, the Traveling Salesman technique solves the
problem [8].
Distributed hash table (DHT) based peer to peer
chord topology is a well-known topology. It main-
tains a finger table for the whole network. For
managing this distributed network, it has used a
chord protocol so that it improves the scalability of
consistent hashing by avoiding the requirement of
having a global network knowledge at every node.
Moreover, in N-node network, the computational
complexity for traversing one node to another node
is O(log N), and also a lookup requires O(log N)
complexity to exchange messages [9].
The M/M/1 queuing model is a well-known and
easily tractable queuing model for the service fulfill-
ment [10]. In several distributed cloud computing
system, the M/M/1 queuing model is widely used which
can efficiently compute the waiting delay. Therefore,
in this paper we have used M/M/1 queuing model.1)
3. System Model
The proposed system model of smart agent-based
dynamic data aggregation for smart city IoT network
is presented in Figure 2. We have proposed a DHT
based peer to peer distributed network for smart city
where every AP is considered as a node and all
other IoT nodes are the elements of the APs. So, the
IoT devices are assigned with one hash key when
these devices connect to the APs. Therefore, we
maintain a finger table for the whole network where
the APs and IoT nodes are equipped with wireless
connectivity. The APs are connected with the BTS
and there is a backhaul connection between the base
transceiver station (BTS) and the core network. In
addition, the cloud is also connected to the core
1) However, in our system model, it is also possible to use M/G/1
queuing model. In that case, the following equation will be used
for calculating the queuing delay[11], m
398 정보과학회논문지 제45권 제4호(2018. 4)
network. The smart agent is a part of the distributed
system where there are multiple agents in the
network and every AP has a queuing model for
serving the request fulfillment.
4. Problem Formulation and Algorithm
We assume the APs are in a chord network and
all the APs of the smart city network follow the
M/M/1 queuing model for request fulfillment. In this
scenario, we focus on the QoS of service fulfillment
by improving the APs selection model for IoT node
and thus, minimizing the queuing delay for the pay-
load transfer. We have considered the APs selection
by using a modified finger table successor and
predecessor based value that takes the maximum
utility so that the network cost becomes minimum
between one AP to another AP. The minimum
network cost APs selection value generates a new
dynamic topology graph to find the minimum cost
APs for a service fulfillment request. We also consider
two types of delay in our model to minimizing the
network cost. First, we consider the AP to AP
traversing delay and second, the waiting delay for
the service fulfillment request from the APs.
Figure 3 depicts a simplified form of the smart city
network where, and are two service providers
with the smart agent and to are the APs. If
the smart agent from service provider requests to
AP for a service, the traverse time between
and becomes the traversing delay and also AP
has some waiting delay for the request fulfillment.
The queuing delay at each AP depends on the
average data transfer volume rate, service rate
and request rate from the service provider .
The summary of the notations is described in
Table 1.
We label the APs with the numbers 1, 2, …, n and
define as,
if
where ∀∈∀∈
is a binary indicator variable where
indicates the request is assigned to and
traverse through the AP to AP , and
otherwise.
Table 1 Summary of Notation
Notation Description
Set of the smart agent
Set of the APs
Requested service provider index
Destination AP index
Source AP index
Traverse delay from to
Processing time in
Request arrival rate from
Service rate from to
Upper bound for waiting time of requests
Data volume transfer rate
Distance cost from to
Total travel time from to per unit time
Total waiting time from to per unit time
For the n numbers of AP in the network, the
distance cost from AP to AP is . Where
… and ≠.
∀∈ ∀∈ ≠ (1)
Equation (1) is the total traversing cost for a
service fulfillment at AP for all possible active
APs form AP to AP .
The service rate from AP to AP is,
.
Here, and are the data volume transfer rate at
and processing time at respectively. Aggregated
service rate at is ∀∈ ≠
.
(2)
In equation (2), we calculate the expected waiting
time at. .
The total waiting time for the requests from the
per unit time is,
ϵ
ϵ
(3)
The total traverse time for requests from the
per unit time is,
ϵ
ϵ
(4)
So, the total delay is the summation of total waiting
time and total traverse time as follow,
지연에 민감한 스마트 시티 서비스를 한 스마트 에이 트 기반의 동 데이터 수집 기법 연구 399
ϵ
ϵ
(5)
The simplified equation of total delay for the
service fulfillment at AP is,
ϵ
ϵ
(6)
The network utility, which is the inverse function
of total cost as follows,
∀∈ ∀∈ ≠ (7)
The total utility gain, for the service request, is
given at equation (7).
ϵ
ϵ
(8)
Equation (8) represents the sum of total utility
for the service fulfillment request.
The objective of the optimization problem in (9) is
to maximize the total network utility.
max ∀∈ ∀∈ (9)
Subject to:
ϵ
ϵ
∀∈ ∀∈ (10)
ϵ
ϵ
≤ ∀ ϵ (11)
ϵ ∀ ϵ ∀ ϵ ≠ (12)
From equation (10) to equation (12) are the constraints
of the objective function (9). Equation (10) has the
guarantee that the request from a smart agent is
already sent to one AP and also every active AP
must have participated in the utility calculation from
the requested AP. Equation (11) determines the esti-
mated waiting time for any service request does not
exceed and in our formulation, we have assumed
that the value of is infinity. Finally, the last
constraint in equation (12) defined the binary decision
making of . The problem in (9) is an NP-hard
and integer linear programming problem. The centralized
solution of the problem (9) leads to large delay and
therefore, not applicable for the real-time IoT app-
lications.
In order to, solve the problem in (9), in this paper,
we propose a heuristic approach to solve this problem.
Firstly, a smart agent finds the APs list for candi-
date nodes from the chord lookup table and calculates
the all possible minimum traversing cost for APs by
using the assigned successor and predecessor APs.
Algorithm 1 Service Fulfillment
1: APList ← Finger table information for
requested APs
2: for each requested AP List
3: for each possible path
4: Initialize constraints using equation
(10),(11),(12)
5: Delay ← solving equation (6)
6: Utility ← solving equation (8)
7: Compute optimal APs using equation (9)
8: end for
9: Service fulfillment completion
10: end for
Besides, it calculates the minimum APs traversal cost
and at the same state smart agent also computes the
expected delay for those APs. Secondly, the smart
agent takes a decision about the maximum network
utility APs that leads to minimum cost for service
fulfillment. Finally, a smart agent fulfills the request
by traversing the minimum cost APs. Alg. 1 presents
the major steps of the proposed smart agent-based
service fulfillment algorithm.
5. Performance Evaluation
We have evaluated the performance of the proposed
approach using simulation in java platform. We have
also used python platform for the result analysis.
Figure. 4 represents the user interface that is deve-
loped and used to evaluate the performance of the
proposed smart agent algorithm.
In our simulation scenario, a chord network is
implemented by using open chord APIs’ [12], and the
smart agent minimum cost AP findings algorithm,
M/M/1 queuing model and the service fulfillment
module are implemented in java platform. We create
the Smart city peer to peer network where the number
of APs and IoT sensor nodes are 51 and 2601
respectively. We have simulated the result for 1 to
30 service fulfillment request parallelly. Figure 5 is a
typical visual description of a service fulfillment that
green line is the minimum costing APs of the service
by using proposed model.
400 정보과학회논문지 제45권 제4호(2018. 4)
Fig. 4 Smart Agent Application
Figure 6 describes the service fulfillment execution
time RTT (round-trip time) comparison for 1 to 30
service request at a time. The dot line (red) is an
existing sink node based centralized system, diamond
shape (blue) is for smart agent based peer to peer
system that is only considered the AP to AP traver-
sing minimum cost (maximizing utility) and rectangle
line (green) combined minimum cost (maximizing utility)
for AP to AP traversing and queuing delay. We get
Fig. 5 Service Fulfillment APs in Smart City Network
Fig. 6 Round-trip time (RTT) for Service Fulfillment
Fig. 7 Convergence Result Previous vs Proposed System
Fig. 8 Waiting Time vs. Utility
a better result when multiple service fulfillment request
s work together by using total utility.
Figure 7 shows the convergence time assessment
for three different cases where it shows that to fulfill
the 10 different service request our proposed model
converge with in 100 micro-second. However, others
two 1000 and 10000 micro-second respectively. We
can perceive that by the increasing number of request
it takes less convergence time compared with sink
node based model and travelling cost minimization
model.
Figure 8 presents the relation between the waiting
time and utility gain. Box plot from Figure 9
describes the mean value of the utility within a
range for three different analysis of the smart city
network.
The average time complexity of our solution is
O((2n)) and the optimal solution for a sink node
based solution is around O(2nn2) and in each
subproblem for this is O(2nn).
지연에 민감한 스마트 시티 서비스를 한 스마트 에이 트 기반의 동 데이터 수집 기법 연구 401
Fig. 9 Box Plot for Utility
6. Conclusion
Smart agent based DDA is a novel approach that
enables a wide variety of smart city services. The
proposed approach substantially helps to reduce the
network overheads by introducing decentralized net-
work management and utility-based AP selection.
Additionally, the computational and dynamic data
aggregation responsibilities are dispersed to multiple
APs instead of single AP. This method will signifi-
cantly reduce the risk of network failure in a dynamic
environment.
References
[ 1 ] A. Zanella, N. Bui, A. Castellani, L. Vangelista, and
M. Zorzi, "Internet of Things for smart cities,"
IEEE Internet Things J., Vol. 1, No. 1, pp. 22-32,
Feb. 2014. (in USA)
[ 2 ] M. G. R. Alam, S. F. Abedin, A. K. Bairagi, A.
Talukder, and C. S. Hong, "An Autonomic SLA
Management for IoT Networks," Proc. of the KIISE
Korea Computer Congress 2016, pp. 507-509, 2016.
(in Korean)
[ 3 ] S. F. Abedin, M. G. R. Alam, R. Haw, and C. S.
Hong, "A system model for energy efficient
green-IoT network," Proc. of the 2015 International
Conference on Information Networking (ICOIN),
pp. 177-182, 2015. (in Cambodia)
[ 4 ] Z. Shen, C. Miao, X. Tao, and R. Gay, "Goal
Oriented Modeling for Intelligent Software Agents,"
Proc. of the Intelligent Agent Technology, 2004,
pp. 540-543, 2004. (in China)
[ 5 ] l. Ganchev, Z. Ji, and M. 0' Droma, "A generic IoT
architecture for smart cities," Proc. of the Irish
Signals & Systems Conference 2014 and 2014 China-
Ireland International Conference on Information
and Communications Technologies (ISSC 2014/
CIICT 2014), pp. 196-199, 2014. (in Ireland)
[ 6 ] M. S. Munir, M. G. R. Alam, and C S Hong,
"Smart Agent Based Data Aggregation for Smart
City," Proc. of the KIISE Korea Computer Congress
2017, pp. 436-438, 2017. (in Korean)
[ 7 ] H. Deng, L Huang, C Yang, H Xu, and B Leng,
"Optimizing virtual machine placement in distributed
clouds with M/M/1 servers," Journal Computer
Communications, Vol. 102, No. C, pp. 107-119, Apr.
2017. (in Netherlands)
[ 8 ] S. H. Rubin, T. Bouabana-Tebibel, Y. Hoadjli, and
Z. Ghalem, "Reusing the NP-Hard Traveling-
Salesman Problem to Demonstrate That P~NP,"
2016 IEEE 17th International Conference on Infor-
mation Reuse and Integration (IRI), pp. 574-581,
2016. (in USA)
[ 9 ] I. Stoica. R. Morris, D. Karger, M. F. Kaashoek,
and H. Balakrishnan, "Chord: A scalable peer-to-
peer lookup service for internet applications,"
SIGCOMM '01 Proceedings of the 2001 conference
on Applications, technologies, architectures, and
protocols for computer communications, pp. 149-160,
2001. (in USA)
[10] Y. Gai, H. Liu, and B. Krishnamachari, "A packet
dropping-based incentive mechanism for M/M/1
queues with selfish users," 2011 Proceedings IEEE
INFOCOM, pp. 2687-2695, 2011. (in China)
[11] T. O. Kim, J. S. Park, H. J. Chong, K. J. Kim and
B. D. Choi, "Performance analysis of IEEE 802.15.4
non-beacon mode with the unslotted CSMA/CA,"
IEEE Communications Letters, Vol. 12, No. 4, pp.
238-240, Apr. 2008.
[12] [Online]. Available: https://sourceforge.net/projects/
open-chord(downloaded 2017, Mar. 04)
Md. Shirajum Munir received the
B.Sc. degree in Computer Science and
Engineering from Khulna University,
Bangladesh in 2010. He served as a
Lead Engineer at solution laboratory,
Samsung R&D Institute, Bangladesh
from 2010 to 2016. Since March 2017,
he is working for his Ph.D. in Computer Science and
Engineering at Kyung Hee University, South Korea. His
research interest includes internet of things, fog
computing, software defined networking, and machine
learning.
402 정보과학회논문지 제45권 제4호(2018. 4)
Sarder Fakhrul Abedin received the
B.S. degree in computer science from
Kristianstad University, Kristianstad,
Sweden, in 2013. Since March 2014, he
has been working toward the Ph.D.
degree in computer science and engi-
neering at Kyung Hee University, Seoul,
South Korea. His research interests include Healthcare
IoT network management, cloud computing, fog computing,
and wireless sensor network. Mr. Abedin is a Member of
the KIISE.
Md. Golam Rabiul Alam (M’17 , S’ 15)
received the B.Sc. degree in Computer
Science and Engineering from Khulna
University and M.S. degree in Inform-
ation Technology from University of
Dhaka respectively. He has been
awarded PhD in Computer Engineering
from Kyung Hee University in 2017. His research interest
includes machine learning, ambient intelligent technologies,
and internet of things.
Do Hyeon Kim received his B.S.
degree in communication engineering
from Jeju National University, in 2014
and his M.S. degree from Kyung Hee
University in 2017. He is currently
working toward his PhD degree at the
department of computer science and
engineering, Kyung Hee University. His research interests
include SDN/NFV.
Choong Seon Hong received the B.S.
and M.S. degrees in electronic enginee-
ring from Kyung Hee University,
Seoul, South Korea, in 1983 and 1985,
respectively, and the Ph.D. degree
from Keio University, Minato, Japan,
in 1997. He served Korea Telecom
Telecommunications Network Laboratory as a Director
of Networking Team from 1988 to 1999. Since 1999, he
has been Professor, Department of Computer Science
and Engineering, Kyung Hee University. His research
interests include future Internet, ad hoc networks,
network management, and network security.