KNOWLEDGE DISCOVERY OF EDGE COMPUTATION FOR

OFFLOAD VEHICULAR APPLICATIONS IN IOT

P.Bhanupriya1, Sabitha Gauni2, K. Kalimuthu3, C. T.Manimegalai4
1
[email protected], [email protected]
1
ResearchScholar, SRM Institute of Science and Technology, Kattankulathur, Chennai
2, 3, 4
Associate Professor, SRM Institute of Science and Technology, Kattankulathur, Chennai

ABSTRACT: One of the biggest challenges in autonomous vehicles is processing massive data
in real-time and task reliable decisions on time. In order to process data in real-time, the onboard unit is
allocated with additional resources, which makes it more complex and consumes much power. To overcome
this, an energy harvesting scheme (EH) along with offloading of tasks is proposed. The main motivation of
autonomous vehicle is to avoid unnecessary accidents caused by negligence and human error. In this paper,
allocation of the tasks between onboard unit and server unit utilizes the resources efficiently, and a novel
task allocation scheme for allocating tasks between onboard unit and server unit is proposed. The decision
of offloading, the processing unit frequencies and the corresponding power transmitted is computed using
the proposed Harvest Energy Residue algorithm (HER). This is a critical feature that enables reliable
communication and produces a greater efficiency in IoT than the existing one. These decisions depend on
direct data obtained, not on the distributed values of the channel, task details, and the EH process. The
model is analyzed for EH and different allocation modes over the entire duration of the task. The results
are simulated and shows that there is an improvement in the ratio of offloading computation tasks.

Keywords: Harvest Energy Residue Algorithm (HER), Energy Harvesting Scheme (EH),
Offloading, Computation, Internet of Things (IoT), Vehicular Applications.

1. Introduction

Due to advancement in recent technologies and its computational capability mobile computing, fog
computing and cloud let becomes more popular in autonomous systems [13]. These system utilize wide
range of sensors for seeding the data from the devices and process the information in real time. As the
demand for multi-function operation of these autonomous system increases, the requirement of huge
number of sensors increases. This brings the system has to collect and process large amount of data which
are limited due to its available resources [14]-[15]. Hence there is a requirement of offloading of these
services to the cloud environment [16]. One of the key challenges is the latency due to offloading of services
which makes it difficult to execute the operation in real time. Hence there exists a tradeoff between seeding
sensor data and the amount of data to be offloaded [18] [20]. A hybrid mechanism to share the computation
among nearby vehicles and infrastructure is proposed in [1]. These mechanism utilizes expensive on board
units for its operation. A vehicle computing resource cloud based autonomous system is designed in [2]
where the some of the computation tasks are shared with nearby vehicles to minimize the computational
delay. Edge computing proven to be successful in handing these computations at offload services. Mobile
edge computing for sharing the edge servers with vehicles resources for better reliability is presented in [3]-
[4]. With this mechanism, multiples tasks are serviced by utilizing multi access of edge servers. And also
by offloading tasks with edge servers reduces latency [5].Sharing the tasks among on board vehicles, edge
servers and cloud is limited and are difficult since a better management of scheduling of tasks with available
resources is needed. By scheduling more tasks to the edge and cloud from onboard unit (OBU) results in
cost effective OBU but requires huge bandwidth. However the network signal is not stable always
especially in dynamic network. On the other hand, by performing much of the tasks in OBU requires high
end OBU units which increases the cost. Moreover the scheduling algorithm must not put additional
overhead on the computation complexity. Another key challenges in offloading is the power required for
transmission. A hybrid power and channel allocation mechanism is proposed in [6] [7]. Most of the
literature discuss about offloading of single user services with some energy harvesting techniques to support
power requirements [8] [17]. For multi user services with multiple access of edge servers, the scenario
becomes worse as it requires robust scheduling algorithms for its reliable operation [9] [19]. In this paper,
an optimal algorithm for offloading computational tasks is proposed. The algorithm considers multi user-
server scenario along with user dynamic behavior model which makes it robust in real world environment.

2. Background
2.1 Internet of Things (IoT)

Computation at the edge of mobile network is possible in Multiple access edge computing (MEC)
only because of Internet of Things (IoT).The most widespread prediction is by 2050 the number of devices
connected to internet will increase five folds more than that of the device users[24]. The wireless standards
used in IoT are RFID, ZIGBEE, GPRS, GSM, Bluetooth and WiFi. The on-board unit (OBU) of vehicle
are connected to the Road Side Unit (RSU), Servers, Base Stations and other devices only because of IoT.
The transport vehicle could become autonomous and the data offloading will be possible only when there
is proper internet connectivity.

2.2 Autonomous Driving Model

An autonomous driving model is shown in Fig. 1. The onboard unit collects the sensory data and
classify the type of object from that data and track that object. These information are mapped against 3D
coordinate system for localization and are then sent for planning the route. The path information is
generated based on behavior and current motion of the vehicle. Later it is transferred to the control unit to
control the actuators connected with mechanical system of the autonomous vehicle.

Fig.1: Autonomous driving model

According to new regulation the autonomous system must track the traffic condition in real time
[11], [24]and respond quickly in less than 100ms [12] ,[25]which is lesser than manual driving system.

2.3 Existing model

The authors in [21] have proposed a model that defines a VANET offloading process. The path of
Vehicle to Vehicle if not available due to network coverage discrepancies then, a necessary path is searched
using WiFi. If the required path not available after search then it is found using cellular network. In the
system model, in order to estimate the vehicle efficiency parameters, smart roadside units having real-time
contact with autonomous vehicles on the road is used. The onboard unit present in the vehicle is borrowed
for V2V communication and macrocell is also involved in computation and offloading services. The
performance of the model is analyzed by calculating the gain difference between the diversified schemes
and a particular offloading scheme, secondly, by calculating the difference of data offloaded by the traffic,
finally, the difference between the messages uploaded by the offloading scheme and the RSU is calculated.

The authors in [22] have done offloading with the help of algorithms namely Heuristic, semi-
definite relaxation (SDR), RLT. Excessive delay causes bottlenecks in the network, preventing data from
filling the pipe and reducing capacity.The greater the latency, the greater the influence on load times. The
difference in latency gain, cost gain with respect to Translatency was done and the effect of offloading on
the cost of the algorithm was analyzed and the results depict 20% latency to be saved in Heuristic algorithm
compared to other algorithm while the cost gain was 74%.

The authors in [23] proposed an energy harvesting model that offloads data in a cost effective way.
Markov decision process (MDP) was used to solve the decision process problem. Lyapunov optimization
based dynamic computation algorithm was used for offloading and it achieves nearly null task drop
compared with mobile execution and dynamic offloading policies.

The authors [24] proposed a genetic offloading scheme which involves crossover and mutation operation.
The optimizing bandwidth and the computation resources can reduce the offloading latency.

The authors in [27] proposed an offloading model that uses Auction algorithm and virtual
machines. The amount of offloading tasks are compared for subscriber’s located 50,70 and 100
meters away from base station and the results show that the offloading was better for shorter
distance. The energy consumption of mobile devices was decreased by 50% in comparison with
the classical Auction algorithm. Parameters like bandwidth and capacity of servers are also
compared between classical auction offloading algorithm and improved auction algorithm. Based
on the comparisons decision model is generated to decide one among the two tasks either to
compute data locally or do data offloading

The object parameter is not processed directly, it is encoded then the process will be performed and
the lowest completion time is considered to be the optimum result. The consumption of energy was
considered while simulating the model.

2.4 Edge Computing Network Architecture

The architecture comprises OBU, mobile edge computing (MEC) server and cloud network. The
OBU is connected with MEC and cloud through wireless edge network and also by means of base station
networks. The wireless edge network comprises of increased data storage and high end device units
integrated with distributed computational units having increased bandwidth capability. The OBU task is
assigned to the wireless edge network when OBU is closer to the wireless edge network. When it is away
from the wireless edge network, The OBU unit access the cloud through cellular network for its services.
It’s shown in Fig. 2.
 Fig. 2: Network Architecture

3. Proposed Model
3.1 System Architecture

The general architecture of the proposed model is shown in Fig. 3. Let width and length of the area
under cover be W andL. The number of OBU be denoted as M and number of MEC is denoted asN. Let t i
be the time taken by the task assigned to OBU unit and t j be the time taken by the task assigned to MEC
unit and the T be the total time T ≥ t i + t j . At a given time slot t the distance between ith OBU units with

the jth MEC unit is given by dti,j < √W 2 + L2 .

Fig. 3: System Architecture

3.2 Computation Model

Let the time taken by the given task to be completed in given time slot t by OBU unit is CTit and
time taken by the given task to be completed in given time t slot by MEC unit is CTjt . Let p be the probability
of assigning a given task to a particular unit either OBU or MEC unit where 0 < p < 1. Before assigning
the computation task to a computational unit at a given time t , the energy residue and the availability of
the time slot and bandwidth is checked. If the unit has poor resource facility, the task gets dropped which
t t
is denoted by Tdrop and thus, CTit + CTjt + Tdrop = 1.

3.3 Offload Model

Consider the decision vector xt = (x1 , x2 , x3 , x4 , … xN ) = ∑N

i=1 xi where N is total tasks assigned.

t
∑N N t N t
i=1 xi = ∑i=1 CTi + ∑j=1 CTj + Tdrop = 0 or 1 (1)

Harvest Energy Residue algorithm (HER)

1: Initially map the network to locate OBU and MEC unit in the given localized area.
𝑡
2: Calculate the distance 𝑑𝑖,𝑗 < √𝑊 2 + 𝐿2 .
𝑡 𝑡
3: Calculate the residual energy 𝐸𝑖,𝑗 and the harvestable energy 𝐻𝑖,𝑗
4: for each task do
𝑡 𝑡
Obtain residual energy 𝐸𝑖,𝑗 and the harvestable energy 𝐻𝑖,𝑗 and generate location
𝑡
Find min (0, 𝑑𝑖,𝑗 )
𝑡 𝑡
Assign Task to OBU unit if(𝐸𝑖,𝑗 > 𝑟𝑒𝑞𝑢𝑖𝑟𝑒𝑑 𝑒𝑛𝑒𝑟𝑔𝑦 𝑅𝑖,𝑗 )& (𝐵𝑊𝑖 > 𝑟𝑒𝑞𝑢𝑖𝑟𝑒𝑑 𝐵𝑊)
for given time with latency𝜁𝑖
Else
𝑡 𝑡 𝑡 𝑡
Calculate Channel Gain 𝐺𝑖,𝑗 = 𝛾𝑖,𝑗 (𝑑0 ⁄ 𝑑𝑖,𝑗 ) and choose the optimal path 𝑝𝑖,𝑗 = disjoint
𝑡
path between 𝑂𝐵𝑈𝑖 𝑎𝑛𝑑 𝑀𝐸𝐶𝑗 . Calculate harvested energy 𝐻𝑖,𝑗
𝑡 𝑡
Assign Task to MEC unit if(𝐸𝑖,𝑗 > 𝑟𝑒𝑞𝑢𝑖𝑟𝑒𝑑 𝑒𝑛𝑒𝑟𝑔𝑦 𝑅𝑖,𝑗 )& (𝐵𝑊𝑗 > 𝑟𝑒𝑞𝑢𝑖𝑟𝑒𝑑 𝐵𝑊)
𝑡
for given time with latency𝜁𝑗 Else Drop the task 𝑇𝑑𝑟𝑜𝑝

4. Algorithm Comparison
The offloading model is analyzed with different algorithm and the proposed algorithm
gives the optimum battery level. The algorithms considered for analysis were Genetic, EPS Greedy,
Differential Evolution, Greedy along with the proposed HER algorithm. Analyzing the battery energy
level for those algorithm for time slots proved that HER is 90 percent more efficient than other
algorithm, it’s mentioned in the Fig. 4 to Fig. 8.
 Fig. 4: Envolution Battery energy level-Genetic Algorithm. Fig. 5: Envolution Battery energy level- EPS Greedy
Algorithm.

Fig. 6: Envolution Battery energy level-Differential Evolution Fig. 7: Envolution Battery energy level-Greedy Algorithm.
Algorithm.
 Fig. 8: Envolution Battery energy level-HER Algorithm.

5. Simulation Results

The simulation results corresponding to proposed model is discussed below. The model is
computed using MATLAB environment and the time slot for a given task is taken as controlled variable
for different dynamic environment. Bernoulli model is used for initial distribution of nodal unit in a given
t 0 0
localized area. Initial value is assigned Emax = Hi,j where Hi,j is the initial harvested power is for a given
unit.
 Fig. 9: Battery Energy Level

The battery energy level corresponding to each OBU unit is depicted in Fig. 9. It is inferred that
the initial energy corresponding to each OBU unit get added at initial stage and after some time the total
energy gets saturated close to the maximum harvested energy level.

Fig. 10: Average Execution Cost

 The average execution cost corresponding each task by various OBU units is depicted in Fig. 10.
It is inferred that at initial stage is much higher due to scheduling of task for various OBU units which
consumes additional complexity and time for path finding and resource allocation and also due to
insufficient energy available at initial stage. This makes the execution cost goes higher at initial stage and
once the allocated the cost gets reduced as tasks allocated are being competed and as time goes by most of
the OBU units are idle without any tasks.

Fig. 11 shows average ratio of mode chosen for given task. It is inferred that the number of task
dropped during earlier stages are higher due to unavailability of the energy at the initial stage. After
sufficient time slot the execution of task by OBU unit and server carries with minimum task drops. Most
tasks are carried by the server since the allocation of resources to the OBU unit are limited and it carries
tasks that are higher priority for performing real time processing of information.

Fig. 11: Average Ratio of Chosen Mode

 6. Conclusion

The discussion about multi-user and multi-server edge computing system has been done.
The model utilizes harvesting of energy schemes along with mode selection for performing
different tasks among OBU unit and server unit. The HER algorithm schedules various task from
various OBU units and utilizes offloading of task to greater extent to better reduce the latency and
complexity in the system and at the same time it produces higher efficiency and reliable
communication makes a good impact for Vehicular Applications in IoT. The simulation results
also demonstrates that offloading autonomous driving services via edge computing can improve
QoS of autonomous driving and reduce local computing resources, effectively. This work is
intended to be extended in future for vehicle offloading with security to be provided by blockchain
technology.

Declarations

Funding:

Not Applicable.

Conflict of Interest:

The authors declare that they have no conflict of interest. The manuscript was written through
contributions of all authors. All authors have given approval to the final version of the manuscript.
Availability of data and material: Not Applicable

Code availability: Not Applicable

REFERENCES

1. The DARPA Urban Challenge, DARPA, Arlington, VA, USA, 2007.

2. S. Kumar, S. Gollakota, and D. Katabi, “A cloud-assisted design forautonomous driving,” in Proc. 1st
Ed. MCC Workshop Mobile CloudComput. (MCC), 2012, pp. 41–46.
3. D. Sabella et al., “Toward fully connected vehicles: Edge computing for advanced automotive
communications,” 5GAA Autom. Assoc.,München, Germany, Rep., 2017.
4. E. Ahmed and M. H. Rehmani, “Mobile edge computing: Opportunities,solutions, and challenges,”
Future Gener. Comput. Syst., vol. 70,pp. 59–63, May 2017.
5. S. Barbarossa, S. Sardellitti, and P.D. Lorenzo, “Communicating while computing: Distributed mobile
cloud computing over 5G heterogeneous networks”, IEEE Single Processing Magazine, vol. 31, 2014,
pp. 45-55.
6. M. Masoudi, B. khamidehi, and C. Cavdar, “Green cloud computing for multi cell networks”, IEEE
Wireless Communications and Networking Conference (WCNC), San Francisco, USA, 2017.
7. Sabitha, G., Manimegalai, C.T., Karthick Narayanan, S.R. et al. “Reduction of Complexity of On-
Board Embedded Robotic System Processors Using Code Offloading” Wireless Pers
Commun ,97 5089–5098 (2017).
8. Y.Y. Mao, J. Zhang, K. B. Letaief, “Dynamic computation offloading for mobile-edge computing
with energy harvesting devices”, IEEE Journal of Selected Areas Communications, vol. 34, 2016, pp.
3590-3605.
9. X.H. Ge, S. Tu, G.Q. Mao, C.X. Wang, and T. Han, “5G ultra-dense celluar networks”, IEEE Wireless
Communications, vol. 23, 2016, pp. 71-79.
10. Apollo. (2017). Autonomous Driving. [Online]. Available:http://apollo.auto/
11. S. Shalev-Shwartz, S. Shammah, and A. Shashua, “Safe, multi-agent, reinforcement learning for
autonomous driving,” 2016. [Online].Available: arxiv.abs/1610.03295.
12. S.-C. Lin et al., “The architectural implications of autonomous driving: Constraints and acceleration,”
in Proc. ASPLOS, Mar. 2018,pp. 751–766.
13. Muhammad Sameer Sheikh, Jun Liang, Wensong Wang, "Security and Privacy in Vehicular Ad Hoc
Network and Vehicle Cloud Computing: A Survey", Wireless Communications and Mobile
Computing, vol. 2020, Article ID 5129620, 25 pages, 2020.
14. Shehzad Ashraf Chaudhry, "Designing an Efficient and Secure Message Exchange Protocol for
Internet of Vehicles", Security and Communication Networks, vol. 2021, Article ID 5554318, 9 pages,
2021.
15. Mengting Yao, Xiaoming Wang, Qingqing Gan, Yijian Lin, Chengpeng Huang, "An Improved and
Privacy-Preserving Mutual Authentication Scheme with Forward Secrecy in VANETs", Security and
Communication Networks, vol. 2021.
16. A. Tassi, I. Mavromatis, R. J. Piechocki and A. Nix, "Secure Data Offloading Strategy for Connected
and Autonomous Vehicles," 2019 IEEE 89th Vehicular Technology Conference (VTC2019-Spring),
2019, pp. 1-2
17. D. Wu, X. Huang, X. Xie, X. Nie, L. Bao and Z. Qin, "LEDGE: Leveraging Edge Computing for
Resilient Access Management of Mobile IoT," in IEEE Transactions on Mobile Computing, vol. 20,
no. 3, pp. 1110-1125, 1 March 2021.
18. A. E. Bayrak et al., "A System-of-Systems Approach to the Strategic Feasibility of Modular Vehicle
Fleets," in IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 50, no. 7, pp. 2716-
2728, July 2020.
19. Y. Cao, D. Li, Y. Zhang and X. Chen, "Joint Optimization of Delay-Tolerant Autonomous Electric
Vehicles Charge Scheduling and Station” ,IEEE Internet of Things Journal, vol. 7, no. 9, pp. 8590-
8599, Sept. Battery Degradation,in 2020.
20. Xiao Jiang, Peng Li, Bin Li, Yulong Zou, Ruchuan Wang, "Security-Reliability Tradeoff for Friendly
Jammer Aided Multiuser Scheduling in Energy Harvesting Communications", Security and
Communication Networks, vol. 2021, Article ID 5599334, 16 pages, 2021.
21. Z. Ning, X. Wang, J. J. P. C. Rodrigues and F. Xia, "Joint Computation Offloading, Power Allocation,
and Channel Assignment for 5G-Enabled Traffic Management Systems," in IEEE Transactions on
Industrial Informatics, vol. 15, no. 5, pp. 3058-3067, May 2019.
22. J. Liu and Q. Zhang, "Offloading Schemes in Mobile Edge Computing for Ultra-Reliable Low Latency
Communications," in IEEE Access, vol. 6, pp. 12825-12837, 2018.
23. Y. Mao, J. Zhang and K. B. Letaief, "Dynamic Computation Offloading for Mobile-Edge Computing
With Energy Harvesting Devices," in IEEE Journal on Selected Areas in Communications, vol. 34, no.
12, pp. 3590-3605, Dec. 2016
24. Keertikumar M., Shubham M. and R. M. Banakar, "Evolution of IoT in smart vehicles: An overview,"
2015 International Conference on Green Computing and Internet of Things (ICGCIoT), 2015, pp. 804-
809, doi: 10.1109/ICGCIoT.2015.7380573.
25. Raiyn, J. “Performance Metrics for Positioning Terminals Based on a GNSS in Autonomous Vehicle
Networks” Wireless Pers Commun , pp. 114,2020.
26. Kwon, B., Park, J. & Lee, S. “A Target Position Decision Algorithm Based on Analysis of Path
Departure for an Autonomous Path Keeping System.” Wireless Pers Commun, pp. 83, 2015.
27. Sheng, J.; Hu, J.; Teng, X.; Wang, B.; Pan, X. “Computation Offloading Strategy in Mobile Edge
Computing” Information 2019, 10, 191.