Professional Documents
Culture Documents
Abstract-With the rising popularity of Internet-enabled mo In parallel with the growth of mobile computing, the
bile devices, users are increasingly demanding better quality cloud computing paradigm has been extensively known as the
of service (QoS). However, the resources of these devices and next generation computing infrastructure. It provides several
their connectivity levels remain insufficient, even though they are benefits by enabling consumers to use computing resources as
improving, for offering acceptable levels of QoS to users. Cloud
utility (e.g., servers, networks, and storages), platforms (e.g.,
computing infrastructures offer large and scalable resources that
middleware services and operating systems), and softwares
allow shifting the physical location of computation and storage
to the cloud. Nevertheless, the integration of mobile computing
(e.g., application programs) offered by cloud providers (e.g.,
with cloud computing would not guarantee adequate levels of Google, Amazon, and Salesforce) at low cost. Furthermore,
service for mobile users. It rather delivers scalability at the cost cloud computing allows users to consume services in an elastic
of higher delay and higher power consumption on the mobile way and access them in an on-demand fashion. Examples
device. Instead, using local resources based on users geographical of services include mobile web search and location-aware
locations has the potential to improve the performance and QoS marketing.
for mobile users. In this paper, we present and study a centralized
architecture that relies on the concept of local clouds, cloudlets, to Taking into consideration the benefits leveraged by the
leverage the geographical proximity of resources to mobile users improvement of resource allocations and the progress of com
and offer them a better user experience. We use a continuous
puting platforms, and the increasing support of cloud services
time Markov-chain (CTMC) to model the different nodes of
for mobile users, the Mobile Cloud Computing (MCC) concept
the architecture: user nodes, cloudlets, and the main cloud.
is presented as an integration of various cloud computing
We estimate the delay incurred in the proposed architecture
by simulating search engine queries generated by mobile users
services into the mobile environment. For example MCC can
using the CTMC state models. Initial simulation results show that be considered as an additional service in which computational
the usage of a cloudlet-based architecture especially centralized intensive data processing and storage services are accessed
architecture has an efficient gains in terms of latency delay and remotely by a mobile user via wireless networks [2]. Neverthe
synchronisation mechanisms. less, the mobile devices are encountering many challenges in
their resources (e.g., battery life, storage, and bandwidth) and
Keywords
communications (e.g., mobility and security) [3], these limited
resources may significantly hamper the amelioration of quality
Mobile Cloud Computing, Cloudlets, Markov chain. of services.
Authorized licensed use limited to: Birla Institute of Technology and Science. Downloaded on February 11,2022 at 06:43:53 UTC from IEEE Xplore. Restrictions apply.
Moreover, most of MCC applications are created upon the architecture where applications are managed at the component
standard web based cloud architecture with mobility support level. The application components can be distributed among
extensions. In such architecture, mobile users have to access the cloudlets. This work, however, does not take into account
cloud services via the Internet by setting up a connection to communication scenarios between cloudlets. Kashif et al. [7]
a cellular wide area network (WAN) such as 3G. This might introduce a new network architecture that integrates distributed
lead to high latency, expensive service access, and significant and local cloudlets to bring cloud resources much closer to
energy budgets due to devices batteries. Radio and battery end users. The architecture benefits from the advantages of
technologies will improve over time but are still expected to wireless mesh networks in terms of cost, efficiency, rapid
be the bottlenecks in future systems [5]. deployment, self-organization, and low-latency access to cloud
services. LifeNet [12] is an ad-hoc routing method that can
To deal with these challenges, we propose the usage of treat different aspects such as node- mobility, obstructions
small clouds known as cloudlets [6], in accessing mobile cloud and node failures. However, LifeNet does not use the concept
services. A cloudlet is a soft state cloud-based application of cloudlets and does not distinguish between the different
service, that contains a data center virtual box, cache copies types of nodes. Besides, LifeNets main focus is reachability
of data or codes and has the ability of self-management with at the expense of throughput. Amazing store [l3] is a low
faster access control [7]. cost P2P cloud storage system that offers high data availability
In this paper, we propose a centralized cloudlet-based and protection against correlated failures. Huaming et al. [14]
architecture that can be configured to respond to the needs have proposed an analytical model for cloudlet-based ofoading
of mobile users. Our proposed cloudlet architecture provides systems which is characterized by cloud unreachability, long
a framework for managing data caches and uses new routing disconnection time or even small speedup factor, this scheme
algorithms for mobile search engine applications which allow will not benet from reduced application execution time. The
users to find mobile content on websites and show a media shift analysis results provide useful guidance for the design of
toward mobile multimedia. Using cloudlet-based architecture efcient ofoading systems. Fesehaye et al. [15] have investigated
has the potential to reduce the high latency, facilitate data the impact of cloudlets on interactive mobile cloud applications
access and data synchronization between the cloudlets and by using cloudlet networks and services such as file editing,
mobile users. video streaming, and collaborative chatting. But they have not
considered scenarios where a network of cloudlets is used for
The remainder of the paper is organized as follows. Section search engine applications on mobile devices. Also, they did
2 describes related work on the issues of Mobile Cloud not investigate centralized cloudlets architectures.
Computing (MCC) and different cloudlet-based architectures.
Section 3 presents our proposed cloudlet-based architecture, To the best of our knowledge, none of the existing ap
defines its caching algorithm and describes our proposed math proaches suggests a centralized architecture that relies on the
ematical model of latency delay. Section 4 presents numerical concept of local clouds, cloudlets, to leverage the geographical
results. Finally, Section 5 concludes the paper and presents proximity of resources to mobile users and offer them a better
future work. user experience and attempts to ameliorate the synchronization
data between several cloudlets. For this reason, we propose
in this paper a centralized cloudlet architecture that can be
II. BACKGROUND AND RELATED W ORK configured by a cloud provider using new routing algorithms
for mobile search engine applications. This architecture aims
Several researchers have investigated the adoption of mo
to: 1) reduce the high latency; 2) facilitate data access; 3) fa
bile cloud computing. Also, many research works have pro
cilitate data synchronization between the cloudlets and mobile
posed Cloudlet-based architectures [7, 11, 12]. As discussed
users. The architecture provides a framework for managing
by Sakr et al. [8], many issues such as scalability, availability,
data synchronization and offloading certain tasks traditionally
reliability and self-awareness need to be addressed to enable
performed at the cloud onto the cloudlets. We focus in this
and adapt mobile computing to cloud environments. A Mobile
work on the routing algorithm to study its performance and
Cloud (MC) needs to dynamically scale the resources require
efficiency of the architecture. Our goal is to demonstrate the
ments of different mobile devices with the demands of cloud
advantages of using a cloudlet-based architecture and evaluate
based mobile applications, and guarantee a minimum level of
its performance, service reliability, and availability.
availability and quality of service. To take advantage of the
cloud, mobile users need to define and specify their acceptable
III. CLOUDLETS ARCHITECTURE
levels of Quality of Service. However, these requirements are
not enough to satisfy mobile cloud needs for additional aspects In this section, we describe our proposed cloudlet-based
such as mobility, low connectivity, and finite sources of power architecture, which aims to reduce the latency and facilitate
[9]. These aspects might lead to high latency and expensive access to data stored in the cloud by mobile users as opposed to
access to cloud services. The cloudlet-based architecture can the classical architecture., Figure 1 illustrates a classical archi
address and alleviate these issues. Sogata et al. [10] implement tecture, where mobile nodes communicate with the cloud via a
MOCHA cloudlet-based architecture, which aims to improve 3G/4G connection. This communication scheme exhibits many
the response time for face recognition applications. However, disadvantages. For example, the delay of the conununication
the MOCHA architecture does not take into account the between two mobiles located in different geographical regions
possible failure of one or more cloudlets, which can hamper the through 3G networks is higher than the delay of the commu
execution of applications. Verbelen et al. [11] propose a more nication through local cloudlets with high wireless bandwidth
dynamic cloudlet concept where mobile devices in the cloudlet (WIFI). Cloudlets play the role of intermediaries between users
network cooperate together. They also present a new cloudlet and the cloud. They facilitate communication and offtoading
Authorized licensed use limited to: Birla Institute of Technology and Science. Downloaded on February 11,2022 at 06:43:53 UTC from IEEE Xplore. Restrictions apply.
@ Router device Super Cloudlet
:e�:�k
��X � (
Base stations (X �X
M h\
/
' At\.
�I I
/
I '
I / ,\ 11 "
1\
I I / I \\ "
\ I \
Mobile
devices i J, �! \:tJ1 �! �'\fJ
i;; '-b ' �
Fig. 1. Standard MC Communication.
the cloud from some tasks such as synchronization in a Fig. 3. Tree topology for MC Communication.
transparent way for the users. Mobile users do not need to
know where their requests and tasks are executed. Some tasks
might be executed on the main cloud, while others are partially nodes, one cloud node and one mobile node). This
executed on the cloudlets. This partitioning depends on the describes a communication scenario for a mobile user.
availability of data and application on the cloudlets. • The Markov chain is in state i 0,1,2,3,
=for central
To demonstrate and illustrate the importance of cloudlets i
ized architecture where is the level of transmission
and the advantages of cloudlets with regards to latency and i
request. In the tree topology, represents the level of
easy access to cloud services, we propose a centralized the node on the tree.
cloudlet-based architecture in which, mobile users are con
• The processes ;::: are mutually indepen
{Nt(i) , t O}
dent homogeneous Poisson Processes with rate A ;:::
nected to their closest cloudlet(s) through wireless network
(WIFI) connection. Like-wise, cloudlets are connected to each
0
which counts the number of arrival requets and the
other through WIFI.
time that these requests occur in a given time interval.
In this work, we use Continuous-Time Markov Chains Nt(i) is a node at level i at instant t
for a given
(CTMC) to represent and model the different states of mobile request, with N(i + 1) N(i) + 1
= and N(O) 1.
=
Authorized licensed use limited to: Birla Institute of Technology and Science. Downloaded on February 11,2022 at 06:43:53 UTC from IEEE Xplore. Restrictions apply.
then it transmits it to the requesting node. Otherwise, it asks
the central cloudlet (i.e. another secondary cloudlet) to search
the requested data in its super-catalog. If the central cloudlet
does not have the requested data, then it forwards the request
to its siblings and to the general cloud.
C. Latency Expression
a=b=l
Mobile users are registered into the user's catalog of their
primary cloudlet. The catalog contains the mapping between i=2;j=C
each users data and its location. Besides, the central cloudlet a=b=l
periodically requires replicas of each cloudlet catalog to update
its super-catalog. When a mobile node submits a request for (1)
data to its primary cloudlet (resp. secondary/central cloudlet), When the appropriate cloudlet of mobile user goes off then the
this latter looks-up in its catalog. If it has the requested data, transmission rate between each node change immediatly and
Authorized licensed use limited to: Birla Institute of Technology and Science. Downloaded on February 11,2022 at 06:43:53 UTC from IEEE Xplore. Restrictions apply.
first appropriate cloudlet goes off:
o i=j=O
"
, .'
/
'" a=1;b=0
�'-;.
,:
A(1-N),
..... p�12],I), �n.J[3],I) ;,.�
�
. , "
:-"� i=l;j=0
P�[I],O),�[2}J)
\.
: � a=0;b=1
..;.�
\ �� Br(N(i),a),(Nj(i+l),b)= A(2-N) 1>, i=2;a=b=1
P�[2],I),�c[3],I) j E [Ln - 3]
i=2;j=C
a=b=1
(4)
N-4 (
1>=PF(t) IT 1 - (k- IJ-?(t) , and
Where
N-4 ( kk=l (t)
) I(
N-
� . 1> = PF(t) D 1 - Prv , and 1> = } 1 - kPrv(t) ,
P�[2],I),N[ ht
.. I ) I )
'
. N-2 ( (k-I F( ) .
J-: t )
"
and <I>= IT 1 -
PN[3],I),N[3],O) k=l
We know that:
Fig. 6. Markov chain model for centralized architecture.
P r[Xc = Nj,F(3)] Nk,F(3)] , Ic/j =I- k, j, k E
[1 N - 3]
,
Let Ai be the rate of the i-th request transition from mobile centralized architecture which is defined as the time needed
node N [0]
to destination node N
[i] and be the total rate of to send the request from the source node to N[k]
node. By
request transitions scattered in the entire architecture, with the conditioning on the node N[i]
will be the destination node of
only non null one-step transition probabilities being : the request, and using the Laplace Setieljes Tansform (LST)
of Tml,c,
we have the latency delay: For e ;::: let 0
i=j=O
a=b=1 E [e-BTm1,c] (6)
N
i=l;j=O LE [e-BTm1,cIXc= k] Pr [Xc=N(i)]
a=b=1 k=l
N
[ 1
-BLTi
i=2;a=b=1 LE e ;=1 IXc= k Pr [Xc=N(i)]
j E [Ln - 3] k=l k
i=2;j=C N ( AN k P r [X =N(i)]
)
a=b=1 = L >:NB c
k=l +
(3)
We can describe another transition probabilty Br when the Moreover, the results of the expected destination node and
Authorized licensed use limited to: Birla Institute of Technology and Science. Downloaded on February 11,2022 at 06:43:53 UTC from IEEE Xplore. Restrictions apply.
the expected latency delay can be computed as follow: cloud experience. The proposed architecture relies on a cen
DE [e-IIeTml,cJ tralized multi-tier topology. We model the delay latency for
DT;;"l,c _
E
[De-IITeml,c ] the cloudlets architecture using bi-dimensional markov chain.
De 111=0 D 111=0 - D 111=0 The results of the simulations we have conducted and show
E [-Tm1,c] -E [Tm1,c]
= the importance of local clouds, or cloudlets, and show that
the centralized cloudlets cloudltes-based architecture allows
Therefore, the expected latency has the following expression: reducing the high access latency and satisfying the users de
mands. In our future work, we plan to compare this architecture
with a the cloudlets-based architecturethat relies on a ring
E [Tm1,c] (7) topology to study the impact of the topology one on some
mobile applications such as search engine applications. We
The expected destination node is given by: also plan to study the use of hybrid architectures: mutli-tiered
N
topologies consisting of multiple groups of cloudlets organized
into rings. Additionally we plan to test the deployment of
LiP, [Xc N(i)] NP, [Xc N(i)]
= = = (8)
i =l cloudlet architectures using the VM technology with different
mobile applications.
>.NE [Tm1,c]
REFERENCES
IV, NUMERICAL RESULTS
[I] s. Mahadev, "Mobile computing: the next decade", the 1st ACM Work
To simulate the request transfer of mobile search engine shop on Mobile Cloud Computing and Services (MCS '10), New York,
applications, we consider scenarios where a mobile user sends NY, USA, 2010, pp. S-6.
a request to different cloudlets, which may be in operational [2] H. Dinh, C. Lee, D. Niyato, P. Wang, "A survey of mobile
or un-operational status. We evaluate the Latency delay of the cloud computing: architecture, applications, and approaches.", DOl:
1O.1002/wcm.1203, 2011.
user requests using Laplace Setieljes Tansform (LST) and the
[3] M. Satyanarayanan, "Fundamental challenges in mobile computing", the
formulation model of the probability of sending the request to
Sth annual ACM symposium on Principles of distributed computing
the appropriate destination node. In the simulation, we consider (PODC '96), New York, NY, USA, 1996, pp. 1-7.
that the requests are sent according to a Poisson distribution [4] http://www. cisco.com/en/US/solutions/collateraVns341/nsS2S/nsS37/ns70
with rate >..
We use the following values for >.: 13%, 28%, S/ns827/white-paper-c11-S20862.html.
50%, and 75%. As shown in Figure 7, when increases >. [S] E. Koukoumidis, D. Lymberopoulos, K. Strauss, J. Liu, and D.Burger,
gradually, the latency delay increases also progressively. "Pocket cloudlets", SIGARCH Computer Architecture News, 2011,
pp.l71-184.
When the number of cloudlets is small and limited to [6] M. Satyanarayanan, P. Bah!, R. Caceres, and N. Davies, "The Case
10, mobile users encounter difficulty in delivering requests to for V M-Based Cloudlets in Mobile Computing", Pervasive Computing,
their appropriate cloudlets, and the latency delay is high time IEEE, 2009, pp.14-23.
due to : a) the time it takes to establish a network between [7] K. Khan, Q. Wang, and C. Grecos, "Experimental framework of in
cloudlets; b) the registration of mobile user's information in tegrated cloudlets and wireless mesh networks", Telecommunications
Forum (TELFOR), 2012, pp.190-193.20-22.
user's catalog; and c) the current state of cloudlets. When
the number of cloudlets varies between and 10the la20, [8] S. Sakr, A. Liu, D. M. Batista, and M. Alomari, "A Survey of Large
Scale Data Management Approaches in Cloud Environments", Commu
tency delay reduces progressively which shows that there is nications Surveys and Tutorials, IEEE, 2011, pp.311-336.
efficent transmission of requests and transparent mechanisms [9] F. Niroshinie, W. L. Seng, and R. Wenny, "Mobile cloud computing: A
of synchronization between cloudlets. In Figure 7 the choice survey, Future Generation Computer Systems", 2013, Volume 29. Issue
of N = 20as number of cloudlets is very specific which 1. pp 84-106. ISSN 0167-739X.1O.1016/j.future.2012.0S.023.
makes the centralized architecture in stable state. Even when [10] T. Soyata, R. Muraleedharan, C. Funai, K. Minseok, and W. Heinzel
the number of cloudlets is greater than 20,
the latency delay man, "Cloud-Vision: Real-time face recognition using a mobile
remains approximately stable to a same value. This is, because cloudlet-cloud acceleration architecture", Computers and Communica
tions (lSCC), 2012 IEEE Symposium on.vol., no., pp.0000S9-000066,(l-
the centralized architecture has a distributed parallel structure 4 July 2012).
with a three levels of transmission: the primary cloudlet, the e [II] T. Verbelen, P. Simoens, F. De Turck, and B. Dhoedt, "Cloudlets:
central cloudlet, and the general cloud. The rate means the Bringing the cloud to the mobile user", 3rd ACM Workshop on Mobile
probability of having events that interrupt the network. When
e Cloud Computing and Services Proceedings, 2012, p.29-3S.
values are respectively 0.1, 0.3
and 0.75;
we notice that the [12] H. Mehendale, A. Paranjpe, and S. Vempala, S, "Lifenet: a flexible ad
latency delay varies very slightly. hoc networking solution for transient environments", ACM SIGCOMM
Comput. Commun. Rev, 41(4):446-447, Aug, 2011.
Therefore, when a mobile user sends a request to its primary [13] Z. Yang, B.Y. Zhao, Y. Xing, S. Ding, F. Xiao, and D. Yafei, "Amazing
cloudlet, which is un-operational then the request is sent to Store: available, low-cost online storage service using cloudlets", the 9th
other cloudlets in the same level of transmission as the primary international conference on Peer-to-peer systems (IPTPS'10). USENIX
Association. Berkeley. CA.USA.2-2.201O.
cloudlet. This contributes to reducing the latency delay and
facilitating the synchronization of data between mobile users [14] W. Huaming, W. Katinka, and A. Grazioli, "Cloudlet-based mobile
offloading systems: a performance analysis", the 31st Int. Sympo. on
and cloudlets. computer performance, modeling and measurement, Austria 2013.
[IS] D. Fesehaye, Y. Gao, K. Nahrstedt, G.Wang, "Impact of Cloudlets on
V, CONCLUSION Interactive Mobile Cloud Applications", the 16th International Enterprise
Distributed Object Computing Conference, IEEE, 2012, pp.123-132.
In this paper we have presented a cloudlets-based archi
tecture that exploits users proximity to improve their mobile
Authorized licensed use limited to: Birla Institute of Technology and Science. Downloaded on February 11,2022 at 06:43:53 UTC from IEEE Xplore. Restrictions apply.
4.2 ,-------;===,---r---,---,-r====,
--+- 1-;0.13
--+- ,;028
--+- ,;0 5
--+- 1-;0.75
10 20 30 40 50 60 70 80 90 100
Number of cloudlets
(a)
16 =0.3 I
--+-,;0.13
--+-1.;0.28
3.5 --+-,; 0.5
�
--+- A; 0.75
,..
� 25
o
�
C
� ....
ID
rn
...J
1.5 �
1
10 20 30 40 50 60 70 80 90 lC
Number of cloudlets
(a)
3 --+-1.;0.13
16=0.75 I --+-,;0.28
--+-,; 0.5
5 --+-,; 0.75
VJ
� ....
E
o
o 2 � ....
5�
1
r----
o.5
10 20 30 40 50 60 70 80 90 1(
Number of cloudlets
(b)
Authorized licensed use limited to: Birla Institute of Technology and Science. Downloaded on February 11,2022 at 06:43:53 UTC from IEEE Xplore. Restrictions apply.