Integrating The K-Out-Of-N Computing Framework For Energy Efficient and Fault Tolerant in Mobi-Cloud
Integrating The K-Out-Of-N Computing Framework For Energy Efficient and Fault Tolerant in Mobi-Cloud
ABSTRACT
       In Personal mobile devices have gained enormous popularity in recent years. Due to their limited
resources (e.g., computation, memory, energy), however, executing sophisticated applications (e.g.,
video and image storage and processing, or map-reduce type) on mobile devices remains challenging
because it need to maintain power consumption. Due to lacking of power in cloud computing of mobile
devices, if any node becomes failure then, entire network of mobile communication may disordered. In
this solution, mobile devices successfully retrieve or process data, in the most energy-efficient way, as
long as k out of n remote servers are accessible. Through a real system implementation the method
proves the feasibility of the proposed approach. Extensive simulations are demonstrated with fault
tolerance and energy efficiency performance of the framework in larger scale networks The integrate k-
out-of-n reliability mechanism into distributed computing in mobile cloud formed by only mobile
devices. K-out-of-n, a well-studied topic in reliability control, ensures that a system of ‗n‘ components
operates correctly as long as k or more components work. More specifically, we investigate how to store
data as well as process the stored data in mobile cloud with k-out of-n reliability such that: 1) The
energy consumption for retrieving distributed data is minimized; 2) The energy consumption for
processing the distributed data is minimized and 3) Data and processing are distributed considering
dynamic topology changes.
Key word: Personal mobile devices, power consumption, fault tolerances, M-K-Out-Data
Allocation, M-K-out-Data Processing
                                                    11
                         International Journal On Engineering Technology and Sciences – IJETS™
                                                       ISSN(P): 2349-3968, ISSN (O): 2349-3976
                                                                  Volume III, Issue V, May- 2016
pool of resources, including data storage space,               network. They are integrating the k-out-of-n
networks, computer processing power, and                       reliability    mechanism      into     distributed
specialized corporate and user applications.                   computing in mobile cloud formed by only
   Mobile Cloud Computing (MCC) is the                         mobile devices. k-out-of-n, a well-studied topic
combination      of   cloud     computing,     mobile          in reliability control to ensures that a system of
computing and wireless networks to bring rich                  n components operates correctly as long as k or
computational      resources    to   mobile     users,         more components work. More specifically, they
network operators, as well as cloud computing                  are study investigate how to store data as well as
providers. The ultimate goal of MCC is to                      process the stored data in mobile cloud with k-
enable execution of rich mobile applications on                out-of-n reliability
a plethora of mobile devices, with a rich user
experience.       MCC          provides      business              In paper our proposed study framework, a
opportunities for mobile network operators as                  data object is encoded and partitioned into n
well as cloud providers.                                       fragments, and then stored on n different nodes.
   In MCC, there are four types of cloud based                 As long as k or more of the n nodes are
resources, namely distant immobile clouds,                     available, the data object can be successfully
proximate      immobile        computing      entities,        recovered. Similarly, another set of n nodes are
proximate mobile computing entities, and                       assigned tasks for processing the stored data and
hybrid (combination of the other three models).                all tasks can be completed as long as k or more
Giant clouds such as Amazon EC2 are in the                     of the n processing nodes finish the assigned
distant immobile groups whereas cloudlet or                    tasks. The parameters k and n determine the
surrogates are member of proximate immobile                    degree of reliability and different k; n pairs may
computing entities. Smart phone‘s, tablets,                    be assigned to data storage and data processing.
handheld devices, and wearable computing                       The following main contribution of the papers,
devices are part of the third group of cloud-                         Under utilization: Typical deployments
based resources which is proximate mobile                      have very low utilization of total computing
computing entities.                                            capacity. Users want to run only a few
   In this research study the first framework to               applications per computer to obtain better
support     fault-tolerant    and    energy-efficient          response time. As a result, many computers are
remote storage and processing under a dynamic                  under-utilized.
network topology, i.e., mobile cloud. The                             Security: Desktops are of ten managed
research framework aims for applications that                  by individual users and they have to regularly
require energy efficient and reliable distributed
data storage      and processing in          dynamic
                                                          12
                          International Journal On Engineering Technology and Sciences – IJETS™
                                                        ISSN(P): 2349-3968, ISSN (O): 2349-3976
                                                                   Volume III, Issue V, May- 2016
apply    security    patches.      Otherwise,     the        recent years, becoming ever more sophisticated
computers can become vulnerable.                             and capable. As a result, developers worldwide
       Operational costs: The total cost of                 are building increasingly complex applications
ownership can grow rapidly for supporting                    that   require        ever   increasing     amounts   of
increasing numbers of desktops and laptops, and              computational power and energy. They propose
for   upgrading     and    up-     dating   software.        ThinkAir, a framework that makes it simple for
Moreover, these computers may waste power as                 developers       to     migrate    their     smartphone
they are often kept on 24 h.                                 applications to the cloud. ThinkAir exploits the
       The objective of this problem is to find n           concept of smartphone virtualization in the
nodes in V as process n nodes such that energy               cloud and provides method-level computation
consumption for processing a job of M tasks is               offloading. Advancing on previous work, it
minimized.                                                   focuses on the elasticity and scalability of the
                                                             cloud and enhances the power of mobile cloud
                                                             computing by parallelizing method execution
2.RELATED WORKS
                                                             using multiple virtual machine (VM) images.
        In this paper, the authors explained that
                                                             In this paper, they propose ThinkAir, a new
mobile applications are becoming increasingly
                                                             mobile cloud computing framework which takes
ubiquitous and provide ever richer functionality
                                                             the best of the two worlds.ThinkAir addresses
on mobile devices. At the same time, such
                                                             MAUI‘s lack of scalability by creating virtual
devices often enjoy strong connectivity with
                                                             machines (VMs) of a complete smartphone
more powerful machines ranging from laptops
                                                             system    on the cloud,           and removes the
and desktops to commercial clouds. This paper
                                                             restrictions                                          on
presents the design and implementation of
                                                             applications/inputs/environmental             conditions
CloneCloud,     a   system       that   automatically
                                                             that CloneCloud induces by adopting an online
transforms mobile applications to benefit from
                                                             method-level offloading. Moreover, ThinkAir
the cloud. The system is a flexible application
                                                             provides an efficient way to perform on-demand
partitioner and execution runtime that enables
                                                             resource allocation, and exploits parallelism by
unmodified mobile applications running in an
                                                             dynamically creating, resuming, and destroying
application-level virtual machine to seamlessly
                                                             VMs in the cloud when needed.
off-load part of their execution from mobile
                                                                      In this paper [4] describe the mobile
devices onto device clones operating in a
                                                             devices are increasingly being relied on for
computational cloud.
                                                             services that go beyond simple connectivity and
        In this paper, the authors explained that
                                                             require more complex processing. Fortunately, a
Smartphones have exploded in popularity in
                                                             mobile         device         encounters,      possibly
                                                        13
                              International Journal On Engineering Technology and Sciences – IJETS™
                                                            ISSN(P): 2349-3968, ISSN (O): 2349-3976
                                                                       Volume III, Issue V, May- 2016
intermittently, many entities capable of lending                      load and limit hops to both stationary and
it computational resources. At one extreme is                         mobile HPC nodes.
the traditional cloud-computing context where a                       2.1 EXISTING SYSTEM
mobile device is connected to remote cloud
resources maintained by a service provider with                               The existing presents a mathematical
which it has an established relationship. In this                     model for both optimizing energy consumption
paper they consider the other extreme, where a                        and meeting the fault tolerance requirements of
mobile device‘s contacts are only with other                          data storage and processing under a dynamic
mobile devices, where both the computation                            network topology. This paper presents an
initiator and the remote computational resources                      efficient     algorithm   for    estimating   the
are mobile, and where intermittent connectivity                       communication cost in a mobile cloud, where
among these entities is the norm.                                     nodes fail or move, joining/leaving the network.
          In   this        paper      [5]     describe     the
geospatially aware mobile devices, such as                                    The project proposed a first process
smart phones, rely on an architecture that is scheduling algorithm that is both fault-tolerant
power con-strained and processing power and energy efficient. It presents a distributed
limited. The utility of these devices can be protocol for continually monitoring the network
computing (HPC) architectures, thus limiting framework processed through a real hardware
battery drain, allowing access to large data, and implementation and large scale simulations.
   If    n-k   node   fails   before   the   tasks        provide functions that take the stored data as
     completion, the task completion ratio is              inputs.
     less than 1 since single cloud provider
                                                                     Each function is instantiated as multiple
     scenario is higher.
                                                           tasks that process the data simultaneously on
   Same execution cost is applied for both
                                                           different nodes. Nodes executing tasks are
     nodes with fixed power supply and nodes
                                                           processor nodes; we call a set of tasks
     powered by battery.
                                                           instantiated from one function a job. Client
   Multiple cloud provider scenarios with
                                                           nodes are the nodes requesting data allocation or
     different execution cost for same task is
                                                           processing operations. A node can have any
     not considered.
                                                           combination of roles from: storage node,
   Storage of node consideration becomes
                                                           processor node, or client node, and any node can
     failure for data processing.
                                                           retrieve data from storage nodes. It considered
   Optimizing the energy efficient becomes
                                                           Topology Discovery and Monitoring, Failure
     tedious for calculating nodes availability.
                                                           Probability Estimation, Expected Transmission
   Failure of node with temporary and
                                                           Time (ETT) Computation, k-out-of-n Data
     permanent disorder for calculating energy
                                                           Allocation and k-out-of-n Data Processing.
     consumption.
2.2 PROPOSED SYSTEM
                                                                     In general, each node may have different
         The proposed system includes all the
                                                           energy cost depending on their energy sources;
existing system approach which covers multiple
                                                           e.g., nodes attached to a constant energy source
cloud service provider environments with
                                                           may have zero energy cost while nodes powered
different execution cost for same task. In
                                                           by battery may have relatively high energy cost.
general, each node may have different energy
                                                           For simplicity, we assume the network is
cost depending on their energy sources; e.g.,
                                                           homogeneous and nodes consume the same
nodes attached to a constant energy source may
                                                           amount of energy for processing the same task.
have zero energy cost while nodes powered by
                                                           As a result ,only the transmission energy affects
battery may have relatively high energy cost.
                                                           the energy efficiency of the final solution. We
The applications generate data and our leave the modeling of the general case as future
fragments are distributed to a set of storage network is homogeneous and nodes consume
nodes. In order to process the data, applications the same amount of energy for processing the
                                                      15
                        International Journal On Engineering Technology and Sciences – IJETS™
                                                      ISSN(P): 2349-3968, ISSN (O): 2349-3976
                                                                 Volume III, Issue V, May- 2016
same task. As a result, only the transmission                   The framework, running on all mobile
energy affects the energy efficiency of the final        nodes, provides services to applications that aim
solution. But in proposed system, different             to:
execution cost for same task scenario is                     Store data in mobile cloud reliably such
considered which resembles the real time case.                that the energy consumption for retrieving
ADVANTAGES                                                    the data is minimized (k-out-of-n data
   Even If n-k node fails before the tasks                   allocation problem);
  completion, the task completion ratio may be               Reliably process the stored data such that
  equal to 1 since multiple cloud provider                    energy consumption for processing the data
  scenario.                                                   is minimized (k-out-of-n data processing
   Different execution cost is applied for both              problem).
  nodes with fixed power supply and nodes                       The application running in a mobile ad-
  powered by battery.                                   hoc network may generate a large amount of
   Multiple cloud provider scenarios with              media files and these files must be stored
  different execution cost for same task is             reliably such that they are recoverable even if
  considered.                                           certain nodes fail. At later time, the application
   Improve a mathematical model for both               may make queries to files for information such
  optimizing energy consumption and meeting             as the number of times an object appears in a set
  the fault tolerance requirements of data              of images. Without loss of generality, we
  storage and processing under a dynamic                assume a data object is stored once, but will be
  network topology.                                     retrieved or accessed for processing multiple
   An efficient algorithm for estimating the           times later. They first define several terms. As
  communication cost in a mobile cloud, where           shown in Fig. 1, applications generate data and
  nodes fail or move, joining/leaving the               our framework stores data in the network. For
  network.                                              higher data reliability and availability, each data
 Scheduling algorithm that is both fault- is encoded and partitioned into fragments; the
 A distributed protocol for continually nodes. In order to process the data, applications
monitoring the network topology, with provide functions that take the stored data as
                                                   16
                        International Journal On Engineering Technology and Sciences – IJETS™
                                                      ISSN(P): 2349-3968, ISSN (O): 2349-3976
                                                                 Volume III, Issue V, May- 2016
                                                    17
                          International Journal On Engineering Technology and Sciences – IJETS™
                                                        ISSN(P): 2349-3968, ISSN (O): 2349-3976
                                                                   Volume III, Issue V, May- 2016
s2; . . . sn ; S V such that the total expected           coding, it first needs to retrieve and decode k
transmission cost from any node to its k closest          data fragments because nodes can only process
storage nodes —           in terms of ETT—is              the decoded plain data object, but not the
minimized. We formulate this problem as an                encoded data fragment.
ILP in Eqs. (1), (2), (3), (4), and (5).                         In general, each node may have different
                                                          energy cost depending on their energy sources;
                                                          e.g., nodes attached to a constant energy source
                                                          may have zero energy cost while nodes powered
                                                          by battery may have relatively high energy cost.
                                                          For simplicity, we assume the network is
                                                          homogeneous and nodes consume the same
                                                          amount of energy for processing the same task.
                                                          As a result, only the transmission energy affects
                                                          the energy efficiency of the final solution. They
                                                          leave the modeling of the general case as future
        The first constraint (Eq. (2)) selects            work. Before formulating the problem, we
exactly n nodes as storage nodes; the second              define some functions: (1) f1(i) returns 1 if node
constraint (Eq. (3)) indicates that each node has         i in S has at least one task; otherwise, it returns
access to k storage nodes; the third constraint           0; (2) f2(j) returns the number of instances of
(Eq (4)) ensures that jth column of R can have a          task j in S; and (3) f3(z; j) returns the
non-zero element if only if Xj is 1; and                  transmission cost of task j when it is scheduled
constraints (Eq (5)) are binary requirements for          for the zth time. We now formulate the k out of
the decision variables.                                   n data processing problem.
                                                                   The objective function minimizes the
3.2 FORMULATION OF K-OUT-OF-N                             total transmission cost for all processor nodes to
DATA PROCESSING                                           retrieve their tasks. L represents the time slot of
        The objective of this problem is to find n        executing a task; i is the index of nodes in the
nodes in V as processor nodes such that energy            network; j is the index of the task of a job. We
consumption for processing a job of M tasks is            note here that T r, the Data Retrieval Time
minimized. In addition, it ensures that the job           Matrix, is a N M matrix, where the element T r
can be completed as long as k or more                     ij corresponds to the estimated time for node i to
processors nodes finish the assigned tasks.               retrieve task j. T r is computed by summing the
Before a client node starts processing a data             transmission time (in terms of ETT available in
object, assuming the correctness of erasure
                                                     18
                       International Journal On Engineering Technology and Sciences – IJETS™
                                                     ISSN(P): 2349-3968, ISSN (O): 2349-3976
                                                                Volume III, Issue V, May- 2016
D) from node i to its k closest storage nodes of         3.3   K-OUT-OF-N DATA ALLOCATION
the task.                                                 After the ETT matrix is computed, the k-out-
                                                         of-n data allocation is solved by ILP solver. A
                                                         simple example of how the ILP problem is
                                                         formulated    and     solved   is   shown    here.
                                                         Considering Fig. 2b, distance Matrix D is a 4 4
                                                         symmetric matrix with each component Dij
                                                         indicating the expected distance between node i
                                                         and node j. Let‘s assume the expected
                                                         transmissions times on all edges are equal to 1.
                                                         As an example, D23 is calculated by finding the
        The first constraint ensures that n nodes
                                                         probability of two possible paths: 2 ! 1 ! 3 or 2 !
in the network are selected as processor nodes.
                                                         4 ! 3. The probability of 2! 1 ! 3 is 0:8 0:8 0:9
The second constraint indicates that each task is
                                                         0:4 ¼ 0:23 and the probability of 2! 4! 3 is 0:8
replicated k - 1 times in the schedule such that
                                                         0:6 0:9 0:2 ¼ 0:08.
any subset of k processor nodes must contain at
least one instance of each task. The third
                                                                Another possible case is when all nodes
constraint states that each task is replicated at
                                                         survive and either path may be taken. This
most once to each processor node. The fourth
                                                         probability is 0:8 0:8 0:6 0:9 ¼ 0:34. The
constraint ensures that no duplicate instances of
                                                         probability that no path exists between node 2
a task execute at the same time on different
                                                         and node 3 is (1-0.23-0.08-0.34 ¼ 0.35). They
nodes. The fifth constraint ensures that a set of
                                                         assign the longest possible ETT ¼ 3, to the case
all tasks completed at earlier time should
                                                         when two nodes are disconnected. D23 is then
consume lower energy than a set of all tasks
                                                         calculated as 0:23 2 þ 0:08 2 þ 0:34 2 þ 0:35 3 ¼
completed at later time. In other words, if no
                                                         2:33. Once the ILP problem is solved, the binary
processor node fails and each task completes at
                                                         variables X and R give the allocation of data
the earliest possible time, these tasks should
                                                         fragments. In our solution, X shows that nodes
consume the least energy.
                                                         1-3 are selected as storage nodes; each row of R
                                                         indicates where the client nodes should retrieve
                                                         the data fragments from. For example, the first
                                                         row of R shows that node 1 should retrieve data
                                                         fragments from nodes 1 and 3.
                                                    19
                        International Journal On Engineering Technology and Sciences – IJETS™
                                                      ISSN(P): 2349-3968, ISSN (O): 2349-3976
                                                                 Volume III, Issue V, May- 2016
                                                   20
                         International Journal On Engineering Technology and Sciences – IJETS™
                                                       ISSN(P): 2349-3968, ISSN (O): 2349-3976
                                                                  Volume III, Issue V, May- 2016
                                                        21
                        International Journal On Engineering Technology and Sciences – IJETS™
                                                      ISSN(P): 2349-3968, ISSN (O): 2349-3976
                                                                 Volume III, Issue V, May- 2016
                                                                           AVG       80
        nt)     Node       Node                Node                     Selection    60                            Selection of
                                                                         Storage                                   Storage MCP
                                                                                     40
                                                                        Node [%]                                   Node
 1      100      16         22      13          17                                   20                            Selection of
                                                                                                                   Storage SCP-
                                                                                         0                         Node
 2      200      32         35      29          30                                           1 2 3 4 5 6 7 8 9
                                                                                             Number of Storage
 3      300      42         49      38          44                                                Node
 4      400      56         62      50          56
 5      500      63         72      56          65          Figure 4.1 Selection Storage Node MEP-SCP
 6      600      71         83      66          80                              Model
                                                              The Table 4.2 represents experimental result
 7      700      79         89      70          85
                                                            for Single K-Out-N-Algorithm and M-K-Out-N-
 8      800      85         92      78          86
                                                            Algorithm mode time analysis. The table
Table 7.1 Selection of Storage Node Analysis contains Data storage node allocation time
finding Data storage node count in K-Out-N and S. Data K-OUT-N M-K-
1 30 20 25
                                                       23
                       International Journal On Engineering Technology and Sciences – IJETS™
                                                     ISSN(P): 2349-3968, ISSN (O): 2349-3976
                                                                Volume III, Issue V, May- 2016
5 70 60 65 details as shown.
6          80             70              75
                                                                                          Data storage Node- Time Analysis
7          90             80              85
                                                                                  140
8         100             90              95                                      120
                                                                  AVG Time[sec]
                                                                                  100
9         110            100             105                                       80                                        K-OUT-N
                                                                                   60                                        M-K-OUT-N
10        120            110             116                                       40
                                                                                   20
Table 4.2 Data storage Node- Time Analysis                                          0
                                                                                        1 2 3 4 5 6 7 8 9 10 11
                                                                                               No.of.Storage Node
       The Fig 4.2 represents experimental
result for Single K-Out-N-Algorithm and M-K- Fig 4.2 Data storage Node- Time Analysis
                                                       24
                   International Journal On Engineering Technology and Sciences – IJETS™
                                                 ISSN(P): 2349-3968, ISSN (O): 2349-3976
                                                              Volume III, Issue V, May- 2016
monitoring and application design. The new      [2]. [Tolia 2006] N. Tolia, M. Kaminsky, D. G.
system     become      useful        if    the   below         Andersen, and S. Patil. An architecture for
enhancements are made in future.                               Internet data transfer. In NSDI, 2006.
 The application can be web service oriented                  [3]. L. Zhang, B. Tiwana, Z. Qian, Z. Wang, R.
   so that it can be further developed in any                  P. Dick, Z. Mao, and L. Yang. Accurate online
   platform.                                                   power     estimation     and   automatic     battery
 The application if developed as web site can                 behavior based power model generation for
   be used from anywhere.                                      smartphones.       In      Proc.     Int.     Conf.
 The algorithm can be further improved so                     Hardware/Software        Codesign    and     System
   that cost of the path can be further reduced                Synthesis, 2010.
reconfigured the storage node Extreme Scale Wireless Sensor Network,‖ Proc.
 Currently the scheme has a slightly less 11th IEEE Int‘l. Conf. Embedded Real-Time
memory overhead, while in the more Comp. Sys. Apps., Aug. 2005, pp. 102–8.
utilize more memory. The future study can Sensor Network System for Energy-Efficient
be in the area of more significant memory Surveillance,‖ ACM Trans. Sensor Net., vol. 2,
                                                          25
                     International Journal On Engineering Technology and Sciences – IJETS™
                                                   ISSN(P): 2349-3968, ISSN (O): 2349-3976
                                                                Volume III, Issue V, May- 2016
BNetwork      coding for distributed storage      http://arxiv.org/abs/1004.4299.
systems,[ IEEE Trans. Inf. Theory, vol. 56, no.            [16]. J. Li, S. Yang, X. Wang, and B. Li,
9, pp. 4539–4551, Sep. 2010.                               BTree-structured         data      regeneration      in
[11].     Y. Wu, A. G. Dimakis, and K.                     distributed storage systems with regenerating
Ramchandran,         BDeterministic   regenerating         codes,[ in Proc. IEEE
codes for                                                  [17]. S. Pawar, S. El Rouayheb, and K.
distributed storage,[ in Proc. Allerton Conf.              Ramchandran, BOn security for distributed
Control Comput. Commun., Monticello, IL,                   storage systems,[ in Proc. IEEE Int. Symp. Inf.
Oct. 2007.                                                 Theory, Jun. 2010
[12]. R. Dougherty, C. Freiling, and K. Zeger,             [18].    cloudstack      -    open     source     cloud
BInsufficiency of linear coding in network                 computing. http://www.cloudstack.org/
information flow,[ IEEE Trans. Inf. Theory,                [19].      IETF       ALTO         working      group.
vol. 51, no. 8, pp. 2745–2759,                             http://datatracker.ietf.org/wg/alto/.
Aug. 2005.                                                 [20]. Windows Remote Desktop Services.
[13]. [33] K. V. Rashmi, N. B. Shah, P. V.                 http://tiny.cc/zt05bw.
Kumar,       and     K.   Ramchandran,       BExact        [21]. M. D. Kristensen, ―Execution plans for
regenerating codes for distributed storage,[ in            cyber foraging,‖ in Proceedings of the 1st
Proc.     Allerton     Conf.    Control   Comput.          workshop on Mobile middleware: embracing
Commun., Urbana, IL,                                       the     personal     communication      device,    ser.
Sep. 2009.                                                 MobMid ‘08. New York, NY, USA: ACM,
[14]. C. Suh and K. Ramchandran, BExact                    2008,     pp.      2:1–2:6.     [Online].    Available:
regeneration codes for distributed storage repair          http://doi.acm.org/10.1145/1462689.1462692
using interference alignment,[ in Proc. IEEE
                                                           [22]. G. F. Carey, E. Barragy, R. McIay, and M.
Int. Symp. Inf. Theory, Jun. 2010. [Online].
                                                           Sharma, ―Element-by- element vector and
Available: http://arxiv.org/abs/ 1001.0107v2.              parallel computations,‖ Comm. Appl. Num.
                                                           Methods, no. 4, pp. 299–308, 1988
[15]. V. Cadambe, S. Jafar, and H. Maleki,
BDistributed data storage with minimum                     [23] Z. Han, A. Swindlehurst, and K. Liu,
storage      regenerating      codesVExact      and        ―Smart deployment/movement of unmanned air
functional repair are asymptotically equally               vehicle to improve connectivity in manet,‖ in
efficient,[ in Proc. IEEE Int. Workshop                    Wireless Communications and Networking
Wireless Netw. Coding, Apr. 2010. [Online].                Conference, 2006. WCNC 2006. IEEE, vol. 1,
Available:                                                 april 2006, pp. 252 –257.
                                                      26
                  International Journal On Engineering Technology and Sciences – IJETS™
                                                ISSN(P): 2349-3968, ISSN (O): 2349-3976
                                                           Volume III, Issue V, May- 2016
[24] R. Hunjet, A. Coyle, and M. Sorell,
―Enhancing mobile adhoc networks through
node placement and topology control,‖ in
Wireless Communication Systems (ISWCS),
2010 7th International Symposium on, sept.
2010, pp. 536 –540.
[25]. G. Karypis, ―Multi-constraint mesh
partitioning for contact/impact computations,‖
in Proceedings of the 2003 ACM/IEEE
conference on Supercomputing, ser. SC ‘03.
New York, NY, USA: ACM, 2003, pp. 56–.
[Online].                            Available:
http://doi.acm.org/10.1145/1048935.1050206
[26]. J. Huang, Q. Xu, B. Tiwana, Z. M. Mao,
M.   Zhang,   and     P.   Bahl.   Anatomizing
Application   Performance     Differences   on
Smartphones. In Proc. of the 8th International
Conference on Mobile Systems, Applications,
and Services (MobiSys), San Francisco, CA,
June 2010.
27