Efﬁcient and Selective Upload of Data from Connected Vehicles

Zaryab Khan

1,2

and Christian Prehofer

DENSO Automotive Germany, Germany

Technical University of Munich, Germany

Keywords:

Connected Vehicles, Data Collection, Connected Mobility.

Abstract:

Vehicles are evolving into a connected sensing platform, generating enormous amounts data about themselves

and their surroundings. In this work, we focus on the efﬁcient data collection for connected vehicles, exploit-

ing the fact that the context data of cars on the same road is often redundant. This is for instance relevant for

applications which need roadside data for map updating. We propose a vehicular data dissemination archi-

tecture with a central coordination scheme to avoid redundant uploads. It also uses roadside WiFi hotspots

opportunistically. To evaluate the beneﬁts, we use the SUMO simulator to benchmark our results against a

baseline solution, showing improvements of factor 10 up to 20.

1 INTRODUCTION

Current vehicles are evolving into a sensing platform,

generating a large amount of real-time data about the

vehicles and their surroundings with cameras, radars

and other sensors. This has led to a huge demand

for data-driven mobility services ranging from traf-

ﬁc management, predictive maintenance, smart cities

to driving behavior analysis. It is forecasted that au-

tonomous cars will generate up to 3,600 gigabytes of

data per hour. An example is road surveillance, where

vehicles equipped with different sensors continuously

monitor the environment

In this work, we focus on the upload of vehicle im-

age data, e.g. for automatic map updating. The goal of

this work is to reduce the upload trafﬁc in connected

vehicles over cellular data connections. We aim to

achieve this goal with two techniques: Firstly, by re-

ducing redundant data and, secondly, by opportunisti-

cally using the available roadside wireless infrastruc-

ture to ofﬂoad mobile data to WiFi. We also use dif-

ferent priorities of the collected data to support differ-

ent kinds of services. The redundancies are detected

by a central control point where vehicles upload im-

age metadata. Overall, our vehicular data dissemina-

tion architecture follows the idea of reverse content

delivery networks (Moustafa et al., 2017), yet focuses

speciﬁcally on avoiding redundant data and on efﬁ-

cient upload.

The goal of this work is to analyze the factors

which inﬂuence the performance of our efﬁcient data

collection in simulations. We evaluate performance

using SUMO (Lopez et al., 2018), an open source

trafﬁc simulation suite, to simulate the mobility pat-

terns of vehicles. The models of SUMO are consid-

ered to be close to real world. Furthermore, we sim-

ulate multiple cities as well as the scalability to up

to 100.000 vehicles. We show that the beneﬁts range

from factor 10 up to factor 20.

To our knowledge, there is no comparable work

focusing on this speciﬁc redundancy avoidance with

central control. This paper is a short version of the

thesis in (Khan, 2019).

2 APPROACH

Our approach is motivated by the existing concept

of content delivery networks (CDN) (Saroiu et al.,

2002), which mainly focus on web content. CDNs

use local copies of content to deliver it efﬁciently to

the end user. For vehicles as data sources, reverse

CDN (R-CDN) was recently proposed in (Moustafa

et al., 2017), focusing on different optimizations for

the upload of video streams.

Compared to existing approaches, the main point

of this work is that many services do not require all

pictures from all cars. For instance, updates every few

seconds or minutes may be enough to detect changes

on the roadside or parking lots or similar. Then, the

point is that the same piece of a road may be visited

by many cars in one minute, and one picture from one

Khan, Z. and Prehofer, C.

Efﬁcient and Selective Upload of Data from Connected Vehicles.

DOI: 10.5220/0009790105590566

In Proceedings of the 6th International Conference on Vehicle Technology and Intelligent Transport Systems (VEHITS 2020), pages 559-566

ISBN: 978-989-758-419-0

559

Figure 1: Image upload with Roid R-CDN.

car may be enough.

Our approach is called Roid: An R-CDN based

selective vehicular data dissemination architecture.

We assume that cars continuously upload the meta-

data of pictures, including time and location to the

control server. The server assigns priorities and sends

requests to vehicles. These upload requested the pic-

tures based on priorities. The main goal of this paper

is to quantify and detail the beneﬁts of our approach

in realistic simulations.

The second mechanism which we employ is of-

ﬂoading cellular network trafﬁc to WiFi networks

along the roadside. The reason behind this is that

WiFi technology is typically cheaper than cellular

technology (Fogg, 2018). This means that, depend-

ing on the priority, data is sent on cellular networks

only if no WiFi network was found for a speciﬁc time

period, e.g. 30 seconds.

As illustrated in Figure 1, the Roid architecture

employs a Control Server (CS) between vehicles and

the cloud storage and processing platform. The pur-

pose of the CS is to orchestrate the vehicles based on

the demand of images in the cloud. The ﬁgure shows

a simple case where an image of a vehicle 2 does not

need to be uploaded as a similar one from vehicle 1

already exists.

A key assumption of the Roid architecture is that

the CS is aware of a complete road network that it

overlooks. The participating cars in this road network

are connected to it and continuously transmit meta-

data like location of pictures taken. The road network

is divided into road segments (typically 30-100m in

length).

The CS server aims to have an up to date picture of

every road segment. Thus, the CS server keeps track

of the latest uploaded image for each segment and

identiﬁes needed images. Clearly, for each segment,

a new image is needed every few minutes. Secondly,

the CS assigns priorities to each segment, which is

done stochastically based on normal distribution. On

the vehicle side, the images are locally stored in ﬁve

queues, based on ﬁve demand classes of the images.

Roid also aims to ofﬂoad a signiﬁcant amount of

upload trafﬁc to WiFi. Each road segment is associ-

ated to one of these ﬁve road segment demand classes.

Then, the CS assigns images from this segment the

corresponding demand class. For each class, the ve-

hicles have different policies regarding WiFi uploads.

Road segments belonging to the maximum class re-

quire no waiting for WiFi, while the others will have a

WiFi waiting time before cellular upload is attempted.

3 IMPLEMENTATION AND

EVALUATION

For the evaluation, we decided to use the Simulation

of Urban Mobility (SUMO) trafﬁc simulation suite

(Lopez et al., 2018). Our Roid implementation is built

on top of SUMO simulator.

The activity diagram presented in Figure 2, de-

scribes the ﬂow of activities in the Roid simulation.

We will go over these step by step below.

1. Start Simulation. The simulation is initialized

in the Roid module by TraCI, an API to interface

ongoing, running simulations. The simulation is

initialized with a number of parameters.

2. 1 Unit Step Into Simulation. The simulation

proceeds in cycles of one time unit.

3. Subscribe to Values of New Vehicles in Simu-

lation. In TracCI, it is possible to subscribe to

the properties of vehicles running in a simulation.

This approach is faster than getting a list of all the

vehicles in each time step of the simulation.

4. Loop Over the Simulated Vehicles. The ap-

proach we follow in this implementation is to loop

over all the spawned vehicles in each time step

and then perform speciﬁc Roid operations on each

vehicle, including their communication with the

Control Server. In the following steps, we will be

performing operations on a single vehicle loop of

vehicles as depicted by the dotted box in Figure 2.

5. Vehicle Takes Image. In each time step, we sim-

ulate vehicles taking an image of the road.

6. Vehicle Saves Image Locally. Every image is

ﬁrst stored in the local vehicle storage before it

is uploaded to the cloud.

7. Vehicle Sends Image Metadata to Control

Server over Control Connection. The image ID,

time and road segment ID are transferred to the

server over the control connection.

8. Server Saves the Image Metadata in a Data

Structure. The server keeps track of the image

metadata received by saving it in a dictionary.

9. Image Required. If additional (not yet requested)

images are required for some road segments, we

proceed with the next step.

10. Server Instructs Vehicle to Send Images. If the

Roid Control Server determined that a particular

VEHITS 2020 - 6th International Conference on Vehicle Technology and Intelligent Transport Systems

560

Figure 2: Activity Diagram of Roid Simulation.

image should be saved to the cloud, it requests the

image and includes the demand class for the road

segment.

11. Vehicle Images are Sorted based on Demand.

On the vehicle side, the images are locally stored

in queues for demand class and newly requested

images are added.

12. WiFi Case Distinction. If WiFi is available, only

WiFi is used in this unit time step.

13. Upload Images. The images that are requested

by the Control Server are uploaded using cellular

data or WiFi, following the priority order.

14. Visualize Data. After all the simulation is tra-

versed, the Roid module calls a visualization

module.

3.1 Evaluation

The performance of Roid will be compared to a sim-

ple upload version, where every vehicle uploads as

much data as possible, and as early as possible. We

use the city of Monaco as our standard scenario for

conducting simulations. The reason for choosing

Monaco is because this is the simulation model that

is recommended by the SUMO open source commu-

nity itself and provides a very realistic trafﬁc model

(Codeca and Harri, 2017).

Table 1 shows the parameters used in the standard

scenario for conducting the Roid performance evalu-

ation test.

WiFi Waiting Time describes the waiting time, be-

fore uploading the image using cellular data.

1. Maximum demand image: 0 seconds

2. High demand image: 60 seconds

3. Average demand image: 300 seconds

4. Low demand image: Only WiFi upload.

5. Minimum demand image: No upload.

In the above standard scenario, there exist roughly

2500 WiFi hotspots spread out over the road network.

The majority of these WiFi hotspots are located at

junctions. For priorities, we assign a demand class

to road segments based on a normal distribution.

Figure 3 shows the results of running a simula-

tion in the standard scenario without using Roid. For

the ﬁrst 3000 seconds, there are less than 50 vehicles

running in the simulation. After that there is a surge

in the trafﬁc of vehicles till the simulation reaches a

value of 1000 vehicles. The number of vehicles in the

system roughly stays the same till we reach the 28000

Efﬁcient and Selective Upload of Data from Connected Vehicles

561

Table 1: System Scenario for Evaluating Roid.

Parameter Value

City Monaco

Unit Time Step 5 seconds

Image Capture Frequency 1 Time Step

Time Limit 36000 seconds

Vehicle upper limit 1000

WiFi Waiting time for Average De-

mand Images

300 seconds

WiFi Waiting time for High Demand

Images

60 seconds

Percentage of active WiFi hotspots 100%

Cellular data packet drop rate &

100%

Figure 3: Total uploads in standard scenario without Roid

(in scale of 1e10).

second mark. After that the number of vehicles in the

simulation quickly drop off to almost 0 vehicles.

The surge of trafﬁc at the 3000 second mark char-

acterizes the rising number of vehicles, for instance

in the morning. Shortly after the surge of vehicle traf-

ﬁc, the vehicles are distributed over the simulation be-

cause the vehicles are spawned over different regions

of the map. Finally, the drop off of vehicles towards

the end of the simulation characterizes the decrease in

vehicular trafﬁc after peak trafﬁc hours have elapsed.

As expected, the total data upload over both WiFi

and mobile connections increases drastically when

more vehicles are active in the simulation. Towards

the end of the simulation, there is 5x more data up-

loaded over cellular connection compared to WiFi

connection. This behavior is expected of the naive,

non-Roid approach as mobile connections are avail-

able all over the map whereas WiFi is available only in

selected regions where uploading over WiFi connec-

tion takes priority over cellular uploads. At the end of

the simulation there was roughly 1 * 10

KB of data

uploaded using cellular network and roughly 2 * 10

KB of data uploaded using WiFi hotspots. Figure 4

Figure 4: Total uploads in standard scenario with Roid (in

scale of 1e8).

shows the results with the Roid architecture. Please

note that this shows uploads in terms of 10

, versus

in Figure 3.

As seen in Figure 4, the total data upload using the

WiFi and mobile connections are remaining roughly

the same for the whole duration of simulation. Com-

paring this behavior with the results from the non-

Roid simulation conﬁrms that the Roid architecture is

successful in ofﬂoading data upload from cellular net-

works to WiFi. The total data upload has also drasti-

cally decreased for both the mobile and WiFi connec-

tions, with the reduction of upload in the former being

more signiﬁcant than the reduction in the upload of

the latter. At the 28000 second mark, the total upload

for mobile connection has reduced from 1 * 10

to 5 * 10

KB. The total upload for WiFi connection

has been reduced from 2 * 10

KB to 5 * 10

KB.

In this experiment using Roid has reduced the mo-

bile connection uploads by 95.1 %. This is quite a

drastic improvement in the efﬁciency of cellular data

uploads, which is achieved by minimizing redundant

image uploads and WiFi ofﬂoading. The reduction in

uploads on WiFi connections using Roid is 77%. It

may be argued that as Roid ofﬂoads data upload from

mobile to WiFi connections, there should be an in-

crease in the total amount of WiFi data upload. This is

not the case as the overall upload volume goes down.

Figure 5 presents the average delay in the upload

of images based on the demand class values. As ex-

pected, the images with maximum demand are in-

stantly uploaded to the Control Server using the cellu-

lar connection. Hence the waiting time of 0 seconds.

The high demand images have a waiting time of 60

seconds per picture. The waiting time for average de-

mand images is a little over 80 seconds while that of

low demand images is around 95 seconds.

Notice that initially, there is high demand as very

VEHITS 2020 - 6th International Conference on Vehicle Technology and Intelligent Transport Systems

562

Figure 5: The average delay in uploading images based on

the demand classes.

Figure 6: Pictured road segments not uploaded to the Con-

trol Server.

little roadside pictures are available. Demand then in-

creases again as more vehicles enter at time roughly

50000, which then levels off due to some saturation

of available pictures.

The waiting times speciﬁed for the high demand

and average demand classes respectively are 60 and

300 seconds. The graph however shows less wait-

ing time than speciﬁed in the demand classes because

these images can also be uploaded on WiFi connec-

tion with less delay than speciﬁed by the image de-

mand classes. This reduces the value of the average

delay in upload time for the image classes.

Figure 6 presents the number of road segments

that were pictured by at least one vehicle but still none

of those images were uploaded to the Control Server.

In a non-Roid simulation, all the pictured road seg-

ments are eagerly uploaded to the Control Server and

may reduce the delay if no overload occurs.

Roid however considers the demand class of im-

ages to minimize the cost of using Roid. The demand

class with the highest number of road segment images

Figure 7: Total WiFi and Mobile upload comparison in

Cologne scenario.

not uploaded to the server is the minimum demand

image class with a value of 322. Next comes the low

demand class with 127 road segments not uploaded to

the Control Server, then average, high and maximum

in the speciﬁed order.

Thus, in this simulation setup, images of impor-

tant road segments almost always get uploaded to the

control server while unimportant road segments may

be neglected, and their images not uploaded to the

control server. This strategy minimizes the cost of

using Roid.

For our evaluation, we also considered other

cities. Figure 7 shows the total data uploaded for

a simulation conducted in the city of Cologne with

1000 cars. The total data uploaded using cellular net-

work in the non-Roid simulation was close to 3 * 10

KB. Using Roid, this value drops down to around 2.5

* 10

KB. A reduction of almost 91% of cellular net-

work usage and an 11-fold efﬁciency gain.

In the next experiment we investigate the effects of

the total number of cars running in a scenario to the

efﬁciency achieved by Roid. Figure 8 shows the WiFi

and mobile uploads for 100,000 vehicles running in

the city of Cologne.

Compared to the scenario where only 1,000 vehi-

cles were running at a time in the city of Cologne, the

performance beneﬁt of Roid in this experiment has

increased from of 11x to 17.73x. As the number of

vehicles in a simulation scenario increases, the con-

gestion on the road network increases. As the conges-

tion on road network increases, the average speed of

vehicles decreases, especially at intersections where

queues form. Many WiFi hotspots are located around

these intersections. The stationary vehicles stuck in

trafﬁc near these intersections are able to ofﬂoad more

data from cellular connection to WiFi. This conforms

to the data of average WiFi availability for vehicle

trips as presented in Table 2.

Efﬁcient and Selective Upload of Data from Connected Vehicles

563

Figure 8: Total WiFi and Mobile upload comparison for

100,000 vehicles in Cologne scenario.

Table 2 shows the beneﬁt of Roid with increase in the

number of vehicles.

Table 2: Roid performance in Cologne Scenario.

Max Vehicle

Limit

Mobile Data

Reduction

Efﬁciency

Gain

Avg. WiFi

Availability

1,000 91% 11x 14.4

10,000 91.4% 11.7x 22.5

25,000 92.1% 12.7x 25.8

50,000 93% 14.4x 28.1

100,000 94.3% 17.73x 30.3

In this experiment, we will change the WiFi waiting

time for high demand images and average demand im-

ages and see how the mobile data uploads can be fur-

ther reduced.

Table 3 shows the performance of Roid in six

simulations instantiated with unique values for the

average Demand Image and High Demand Image

WiFi waiting times. The ﬁrst row of the table cor-

responds to the parameters in the standard scenario.

As expected, higher waiting times for image demand

classes result in efﬁciency gains for mobile uploads

while efﬁciency losses for WiFi uploads as more data

is ofﬂoaded from mobile connection uploads to WiFi

uploads.

Table 3: Roid performance with different WiFi Waiting

Times.

Average De-

mand Image

High De-

mand Image

Efﬁciency

Gain Mobile

Efﬁciency

Gain WiFi

300 60 1x 1x

300 120 1.09x 0.93x

600 60 1.10x 0.93x

600 120 1.21x 0.92x

1200 250 1.40x 0.91x

2500 400 1.51x 0.90x

Figure 9 visualizes the experiments with the plots of

Figure 9: Total uploads on mobile connection based on im-

age demand class values.

the total WiFi uploads for the different Image Demand

Wait times. The labels of Figure 9 correspond to the

tuple of average Demand Image and High Demand

Image parameters. As the WiFi wait times increase,

the total uploads decrease. However, the decrease in

total uploads is not signiﬁcant.

Even as the waiting times for Average Demand

and High Demand are increased signiﬁcantly to the

values of 2500 s and 400 s respectively, the efﬁciency

gain achieved for mobile uploads is only 1.51x. That

is not a very signiﬁcant efﬁciency gain considering

an increase in waiting time by factor 8. Similarly,

the WiFi uploads change inversely. One explanation

is that the vehicles in the simulation do not require

a long amount of time to reach WiFi hotspots. We

should also note that increasing WiFi wait time add

additional delay. In our standard Monaco scenario,

high demand images are uploaded on WiFi after an

average delay of roughly 60 seconds while the aver-

age delay for high demand images is roughly 93 sec-

onds. Compared to the standard scenario, the average

delay times for both high demand images and average

demand images are roughly the same at a value of 120

seconds.

In the standard scenario, there exist roughly 2500

WiFi hotpots spread out over the road network. We

evaluate how reducing the number of active WiFi

hotspots increases the amount of data uploaded using

mobile connections.

Table 4 indicates the effect of changing the num-

ber of WiFi hotspots on the total mobile and WiFi up-

loads in Roid.

The interesting result is that with just 30% of the

access points, we ofﬂoad more than 50% or the case

with all 2500 access points.

Another important point is the variability of our

results regarding on different simulation runs. For

the Monaco scenario, variance of results decreases

VEHITS 2020 - 6th International Conference on Vehicle Technology and Intelligent Transport Systems

564

Table 4: Roid performance with decreasing number of WiFi

hotspots.

% active WiFi

hotspots

Increase Mobile

Upload

Increase WiFi Up-

load

100 1x 1x

70 1.09x 0.90x

50 1.21x 0.76x

30 1.37x 0.53x

0 1.78x 0x

strongly when number of vehicles increases, down to

1.9% for 1000 vehicles (Khan, 2019). The reason for

the low variance is the random distribution of vehi-

cles. Randomness is be integrated in the simulation

in a number of ways:

• Randomize vehicle departure times in ﬁxed

routes.

• Randomize the driving behavior of vehicles.

• Randomize the routes of vehicles.

4 RELATED WORK

Overall, we could not ﬁnd any speciﬁc, comparable

work. A main difference is that we assume a cen-

tral coordination server, which we consider realistic

as vehicles upload trafﬁc data anyway, typically for

congestion analysis. In the following, we compare to

similar approaches.

The idea of Dynamic Interrogative Data Capture

(DIDC) is to identify the smallest data collection and

transmission rates to provide information requested

by the transportation authority on different trafﬁc sit-

uations (Wunderlich, 2017). The problem with this

approach is that there may be conﬂicting requests and

the task of prioritization and sorting of the requests is

not trivial (Lin et al., 2018).

Regarding WiFi ofﬂoading, (Lee et al., 2010)

shows that a signiﬁcant amount of mobile data can be

ofﬂoaded for pedestrians using smartphones. Here,

we analyze the ofﬂoading behavior in vehicles that

cover distances much faster than pedestrians and

show similar results for this aspect.

For WiFi trafﬁc other works show that congestion

and transfer latency usually result from poor alloca-

tion of WiFi resources (Hossain et al., 2010). Other

approaches like (Sikdar, 2008), Cabernet (Eriks-

son et al., 2008), Drive-Through Internet (Ott and

Kutscher, ) and Shared Wireless Infostation Model

(SWIM) (Small and Haas, 2003) consider fairness

among users. The above works could complement

our ﬁndings here, but do not consider the ofﬂoading

as such in realistic scenarios.

The general idea of selective data upload has also

been investigated in sensor networks. Data selection

techniques, also known as suppression-based tech-

niques, have been proven useful and accurate in mon-

itoring physical phenomena while signiﬁcantly reduc-

ing the amount of transferred data (Kulik et al., 2008;

Puggioni and Gelfand, 2009; Silberstein et al., 2006).

The most basic method is temporal suppression where

connected vehicle data is only transmitted when the

data is different from what the vehicle and the central

server expect. The main difference is that in our case,

the mobile vehicles continuously transmit metadata to

a central server which coordinates the data collection.

5 SUMMARY

We have presented a new approach, Roid, for reduc-

ing redundant data from connected vehicles uploads,

using a central control architecture for one speciﬁc

road network. We assume that this can be scaled by

using several such servers for multiple regions. As

vehicles increasingly use connected services like nav-

igation and trafﬁc warnings, it is reasonable to assume

that there is a continuous connection for metadata up-

load. The mechanism reduces redundancy based on

locality of pictures, plus WiFi ofﬂoading and priori-

ties to model different services.

We analyze the beneﬁts of different factors like

WiFi upload, as well as the scalability to up to

100.000 vehicles. We show that the beneﬁts range

from factor 10 up to factor 20, and clearly show that

more trafﬁc can improve the beneﬁts. The factor

could also be higher if the available network resources

would be higher, so the simple approach would up-

load even more. We also show that longer waiting

times for WiFi hotspots does not give major improve-

ments in our simulations. On the other hand, we can

show that half of the WiFi upload can be achieved

with just 30% of the access points. Effect of data of-

ﬂoading to WiFi with some wait - analyzed waiting

time and density of WiFi access points.

REFERENCES

Codeca, L. and Harri, J. (2017). Towards multimodal mo-

bility simulation of c-ITS: The monaco SUMO trafﬁc

scenario. In 2017 IEEE Vehicular Networking Con-

ference (VNC). IEEE.

Eriksson, J., Balakrishnan, H., and Madden, S. (2008).

Cabernet. In Proceedings of the 14th ACM interna-

tional conference on Mobile computing and network-

ing - MobiCom 08. ACM Press.

Efﬁcient and Selective Upload of Data from Connected Vehicles

565

Fogg, I. (2018). The state of wiﬁ vs mobile network expe-

rience as 5g arrives.

Hossain, E., Chow, G., C.M.Leung, V., D.McLeod, R.,

Mivsic, J., W.S.Wong, V., and Yang, O. (2010). Vehic-

ular telematics over heterogeneous wireless networks:

A survey. Computer Communications, 33(7):775–

793.

Khan, Z. (2019). Efﬁcient data collection from con-

nected vehicles. Master Thesis, Technische Univer-

sit

at M

unchen, Germany.

Kulik, L., Tanin, E., and Umer, M. (2008). Efﬁcient data

collection and selective queries in sensor networks.

In GeoSensor Networks, pages 25–44. Springer Berlin

Heidelberg.

Lee, K., Rhee, I., Lee, J., Chong, S., and Yi, Y. (2010).

Mobile data off-loading: How much can wiﬁ deliver?

ACM CoNEXT 2010.

Lin, L., Peeta, S., and Wang, J. (2018). Efﬁcient collec-

tion of connected vehicle data based on compressive

sensing. 21st International Conference on Intelligent

Transportation Systems (ITSC).

Lopez, P. A., Behrisch, M., Bieker-Walz, L., Erdmann, J.,

Flotterod, Y.-P., Hilbrich, R., Lucken, L., Rummel,

J., Wagner, P., and WieBner, E. (2018). Microscopic

trafﬁc simulation using SUMO. In 2018 21st Interna-

tional Conference on Intelligent Transportation Sys-

tems (ITSC). IEEE.

Moustafa, H., Schooler, E. M., and McCarthy, J. (2017).

Reverse cdn in fog computing: The lifecycle of video

data in connected and autonomous vehicles. 2017

IEEE Fog World Congress (FWC).

Ott, J. and Kutscher, D. Drive-thru internet: IEEE 802.1

1b for ”automobile” users. In IEEE INFOCOM 2004.

IEEE.

Puggioni, G. and Gelfand, A. E. (2009). Analyzing

space-time sensor network data under suppression and

failure in transmission. Statistics and Computing,

20(4):409–419.

Saroiu, S., Gummadi, K. P., Dunn, R. J., Gribble, S. D., and

Levy, H. M. (2002). An analysis of internet content

delivery systems. ACM SIGOPS Operating Systems

Review, 36(SI):315–327.

Sikdar, B. (2008). Design and analysis of a MAC proto-

col for vehicle to roadside networks. In 2008 IEEE

Wireless Communications and Networking Confer-

ence. IEEE.

Silberstein, A., Braynard, R., and Yang, J. (2006). Con-

straint chaining. In Proceedings of the 2006 ACM

SIGMOD international conference on Management of

data - SIGMOD 06. ACM Press.

Small, T. and Haas, Z. J. (2003). The shared wireless info-

station model. In Proceedings of the 4th ACM inter-

national symposium on Mobile ad hoc networking &

computing - MobiHoc 03. ACM Press.

Wunderlich, K. (2017). Dynamic interrogative data capture

(didc). Concept of Operations — National Operations

Center of Excellence.

VEHITS 2020 - 6th International Conference on Vehicle Technology and Intelligent Transport Systems

566