FogRoute: DTN-based Data Dissemination Model in Fog Computing · FogRoute: DTN-based Data Dissemination Model in Fog Computing Longxiang Gao Member IEEE, ... Fog servers, an important

IEEE Internet of Things Year: 2017, Volume: 4, Issue: 1

1

FogRoute: DTN-based Data Dissemination Modelin Fog Computing

Longxiang GaoMember IEEE, Tom H. LuanMember IEEE, Shui Yu Senior Member IEEE, Wanlei ZhouSeniorMember IEEE, and Bo LiuMember IEEE

Abstract—Fog computing, known as “cloud closed to ground”,deploys light-weight compute facility, called Fog servers, atthe proximity of mobile users. By pre-catching contents in theFog servers, an important application of Fog computing is toprovide high-quality low-cost data distributions to proximitymobile users, e.g., video/live streaming and ads dissemination,using the single-hop low-latency wireless links. A Fog computingsystem is of a three tier Mobile-Fog-Cloud structure; mobile usergets service from Fog servers using local wireless connections,and Fog servers update their contents from Cloud using thecellular or wired networks. This, however, may incur high contentupdate cost when the bandwidth between the Fog and Cloudservers is expensive, e.g., using the cellular network, and istherefore inefficient for non-urgent, high volume contents. Howto economically utilize the Fog-Cloud bandwidth with guaranteeddownload performance of users thus represents a fundamentalissue in the Fog computing. In this paper, we address the issueby proposing a hybrid data dissemination framework whichapplies SDN (software defined network) and DTN (delay tolerablenetwork) approaches in Fog computing. In specific, we decomposethe Fog computing network with two planes, where the cloud is acontrol plane to process content update queries and organize dataflows, and the geometrically distributed Fog servers form a dataplane to disseminate data among Fog servers with delay tolerantnetwork technique. Using extensive simulations, we show thatproposed framework is efficient in term of data disseminationsuccess ratio and content convergence time among Fog servers.

Index Terms—Fog Computing, Delay Tolerant Network, DataDissemination

I. I NTRODUCTION

FOG computing [1], [2] is an emerging technology toprovide high quality engaged localized services. The Fog

computing extends Cloud services by deploying Fog servers atthe physical proximity of mobile users, e.g., parks, shoppingmalls. This provides a local software defined “customizedCloud” accordingly at user’s premises with dedicated cloudwhich possesses processing, storage and network transmissionability. Compared with centralized Cloud services, such asAmazon EC2 and IBM SoftLayer, Fog computing spreadsits facility at distributed locations surrounding users, andaccordingly provides more engaged, localized mobile servicestailored to the deployment sites, e.g., streaming applications,localized advertisements, sensor networks. In order to providethe seamless services in wide areas, Fog servers are geograph-ical distributed and managed by a central cloud controller.They can be widely deployed in remote rural area, city center,

L. Gao, T. H. Luan, S. Yu, W. Zhou and B. Liu are with the School ofInformation Technology, Deakin University, Burwood, VIC 3125, Australia.Email: {longx, toml, syu, wanlei, bo.liu}@deakin.edu.au.

moving vehicles and autonomous systems to support a verylarge number of nodes, which means wireless enabled devicescould use Fog computing technique in anytime, anywhere andanyhow.

Fog computing typically has a three-tier Mobile-Fog-Cloudstructure [3]. In the Mobile tier, it could include all wirelessdevices, such as smartphones, tablets, laptops. In the Fog tier,Fog servers provide services to the end users and synchronizedata with the Cloud. In the Cloud tier, Cloud provider providescontent service required by geo-distributed Fog servers. Datadissemination between a mobile user and a Fog server isoccurred when this mobile user retrieves content. If this Fogserver has the required content, it sends the content to themobile user. Otherwise, this Fog server needs to send a queryto its Cloud provider to find and download it into its localstorage. On another side, Fog servers need to regularly checkwith their Cloud providers whether the Fog servers have theupdated contents or not; if not, they need to update their stor-age by retrieving from the Cloud via either wired or wirelessnetworks, e.g., cellar networks. Such data disseminations mayinvolve a huge cost due to the large data volume. Accordingto the report from Cisco [4], the overall mobile data trafficis expected to grow to24.3 exabytes per month by2019 andmore traffic will be offloaded from cellular networks, such asFog devices, than remains on cellular networks by2016.

The motivation of this paper is the “two-latency” prob-lem in Fog computing. The first “latency” is the time todeliver the demanded content to mobile users. Streamingapplications are becoming more and more popular, while thelong latency may severely affect the user experience and isnot tolerable. To address this issue, Fog computing movesCloud services from remote Internet to the edge of networksand makes streaming content much closer to mobile users,which significantly decreases the streaming latency. On thecontrary, the data dissemination from Cloud to every Fogservers can be expensive, particular for the areas without afast and reliable Internet access. In addition, note that sincethe majority of streaming applications are video based, suchas movies, teleplay and product advertisement, such contentsare not always necessary to be strictly up-to-date and one ortwo days latency is affordable, which is the second “latency”.Consequently, data dissemination for the high volume and non-urgent data among Fog servers becomes promising and delaytolerant network technique makes it available. In summary, thecontributions of this paper are three fold:

• Practical system: we have developed a practical Fogcomputing based content dissemination framework in


2

vehicular networks. The proposal makes use the delaytolerant model and also the Cloud computing framework.

• Distributed algorithm: we have developed a fully dis-tributed algorithm to enable the low-latency and low-costcontent disseminations in the vehicular fog computingsystem.

• Performance evaluation: we have conducted extensivesimulations to verify the performance of the proposal.It is show that our proposal can achieve a low end-to-end delay of data dissemination with high successfulprobability.

The rest of paper is organized as follows. Section II reviewsthe existing models and introduces the background of Fogcomputing. Section III describes the hybrid data disseminationmodel in details. Section IV analyzes simulation performancesof the proposed model. Finally, Section V summarizes thispaper and provides suggestions for future works.

II. BACKGROUND AND RELATED WORKS

A. Fog Computing

Fog computing [3], [5], [6], [7] is an emerging paradigm toextend the Cloud service to the “ground”, where applicationsor contents located on far away worldwide Cloud servers arenow distributed to a multitude of Fog servers located just inlocal business premiss, on-board or even the outdoor, whichcan also be used in urban computing [8]. These Fog servershave process, storage and network transmission capabilities.Mobile users access Fog servers with just one-hop wirelessconnection, and therefore streaming applications for majorityof mobile users become available even at the area in a poorInternet coverage. The transmission speed from Fog server tomobile user is much faster than the speed from remote Cloudto end users. For example, the IEEE 802.11ac WiFi has aspeed around 2600Mbit/s in its 5 GHz band, while the mostlatest 4G LTE Advanced cellular network only have a speedaround 1000Mbit/s.

As Fog computing is a promising paradigm, several papershave discussed its definition, purpose, characters and researchtrend, such as. In this section, we mainly focus on thefollowing real applications or architectures under the umbrellaof Fog computing.

A Fog computing network is managed by a central Cloudserver. This accordingly makes Fog computing a three-tierhierarchy structure, Mobile-Fog-Cloud. There are three bidi-rectional data dissemination modes in this model:Mobileto Fog, Mobile to Cloud and Fog to Cloud. Mobile to Fogdata dissemination is the main purpose of Fog computingparadigm, where majority of contents used in a particularlocation have been pre-downloaded from Cloud to this Fogserver. Mobile users can easily access these resources, suchas using WiFi technique, with a very high transmission speedand a reliable network environment. High quality streamingapplications become available with a low cost. In the caseif a Fog server has not the content a mobile user requested,Mobile to Cloud data dissemination is used. In this mode,end user can directly retrieve content from Cloud provider,which is the same as the current Cloud model. In order

to provide this Fog computing service, Fog servers need toget the up-to-date contents from Cloud service provider byusing theFog to Cloud mode. This data dissemination modeis usually conducted by Internet connection, such as ADSL,cable network and optical network. For these Fog servers undermobile environment or pool broadband coverage areas, cellaror even satellite network is used.

B. Data Dissemination based on Delay Tolerant NetworkTechnique

For high volume and non-urgent data, it is not wise touse expensive network transmission techniques. Instead, delaytolerant network technique could be used to provide datadissemination between Fog servers, and between mobile usersand Fog servers as well.

Delay Tolerant Network (DTN) [9], [10] is featured withlong latency and unstable network topology, where end-to-enddelay can be measured in hours or days and data disseminationpaths may not exist. Data dissemination in DTN uses store-carry-forward techniques; if there is no connection availableat a particular time, the source node needs to store and carrythe message until the next available node is encountered.

DakNet [11] is a DTN technique based application anddeveloped by researchers from the MIT Media Lab. It hasbeen deployed in remote parts of Cambodia and India at acost two orders of magnitude less compared to traditionallandlines networks. DakNet uses existing communicationsand transport infrastructure to form a delay tolerant network,where it combines physical transportation with a wireless datatransferral function to extend Internet connectivity to a centralhub, such as cyberspace or a post office. DakNet does notneed to relay data over a long distance, which is an advantagebecause it would reduce costs and save a lot of power. Instead,DakNet transmits data over short point-to-point links betweenkiosks and portable storage devices called mobile access points(MAPs). A MAP can exchange data among public kiosks,Internet enabled devices and non-Internet accessed hubs byusing mobile generators mounted on and powered by a bus ormotorcycle.

The point-to-point data transfers use high bandwidth withlow-cost WiFi radio transceivers. When a MAP-equipped vehi-cle comes within range of a WiFi-equipped kiosk, it automat-ically detects the wireless connection and uploads/downloadstens of megabytes of data. On the other hand, when the MAP-equipped vehicle comes within range of an Internet accesspoint, it automatically synchronizes the data generated fromrural kiosks. This application creates a low-cost delay tolerantnetwork and asynchronous data exchange infrastructure. Thisprocess would be sufficient to provide the daily informationservices for a small village with only one small vehicle passingeveryday.

As demonstrated above, DTN technique could be used asa low cost data dissemination method in Fog computing,particular for data dissemination among Fog servers andmobile devices. If these become available, data traffic betweenFog server and Cloud could reduce dramatically. In order toimplement it, two main carriers are used in this DTN based


3

data dissemination, human associated DTN [12], [13], [14] andvehicle based DTN [15]. Most of mobile devices, associatedwith Fog servers, are either carried by human or mounted onvehicle. The movement pattern of human carried devices arepredictable and these devices can be utilized to carry small tomedium size content to provide an offload data disseminationin human associated DTN [16]. The movement pattern ofvehicles are classified in the following three categories:

• Fixed Route: Vehicles move to the predefined destinationwith pre-configured routes, such as shuttle bus and train.

• Flexible Route with Fixed Destination: Vehicles move tothe predefined destination with different routes, such astaxi and GPS enabled car.

• Expected Route: Vehicles move along a certain road, suchas the road between two intersections.

III. H YBRID DATA DISSEMINATION IN FOG COMPUTING

As identified above, the current data dissemination modelin Fog computing is not efficient for these high volume andnon-urgent content. In this section, we presentFogRoute, ahybrid data dissemination model using Fog computing, whichtakes advantages of delay tolerant network (DTN) techniqueto offload content among Fog servers either by human orvehicle. The first sub-section introduces the new model tosupport this hybrid data dissemination. The second sub-sectionpresents how the hybrid data dissemination works in Fogcomputing topology, and the last sub-section discusses privacyand reliability issues of FogRoute.

A. System Model

To have an efficient data dissemination in Fog computing,DTN is used to offload data transmitted among Fog servers.In Fig. 1, if Fog server A has some small contents, such as anew video ads, but Fog server B does not. In previous, Fogserver B needs to get the update from Cloud server directly.With DTN technique, user 1 could download this content whenhe has a coffee in this cafe shop. After a couple of hours, hegoes to shopping center for shopping. When he moves into thetransmission range of Fog server B, this content storied in user1’s mobile device automatically upload to Fog server B andthis store-carry-forward process is finished. For large content,e.g. a high definition movie, transmitted from Fog server Bto Fog server A, vehicle based DTN is used. In this example,if car 2 is parked in the shopping center and within the Fogserver B’s transmission range, it downloads this movie intoits local storage. Once this vehicle moves to the transmissionrange of Fog server A, this movie is uploaded to it and thisdata dissemination is completed.

In addition of the above DTN based data disseminationbetween mobile user and Fog server, direct data disseminationfrom Fog server to Fog server based on DTN technique is alsoavailable. Assuming that an on-board Fog server is installed ona tourism bus, so that passengers on this bus can access its Fogserver to watch movies or play games. If a passenger wantsto find new contents, such as “just in” news, Fog server canuse the cellular network to retrieve this content immediately.When this bus travel to a small town along its route, it can

Cloud Server

Location 1

(A Café Shop)

Location 2

(A Shopping Center)

Cloud

Fog

Mobile

Data Dissemination

Fog

Server A

Fog

Server B

User 1 User 2Car 2Car 1

Fig. 1: Data Dissemination in Fog Computing based on DelayTolerant Network Technique

Location 1 Location 2 Location 3 Location 1 Location 2 Location 3

Control Plane

Data Plane

Fog Server Fog Server Fog Server

Data Dissemiantion

ChannelWiFi Channel

Cloud Server

Fig. 2: Hybrid Data Dissemination Model in Fog Computing

synchronize its content with the Fog server located in thissmall town to update both servers’ content list. If the Fogserver in this small town has some content while other smalltowns along the route do not have, this tourism bus coulddownload these contents into its local storage and carry tothose small towns where they need these contents.

With the above DTN based data dissemination techniques,we propose a hybrid data dissemination model, as shownin Fig. 2. This hybrid model not only includes the normaldata dissemination between Cloud servers and Fog servers,but also involves large amount low-cost DTN based datadisseminations, which can be used between Fog servers andmobile users and among Fog servers. To organize these datadissemination, we re-identify the function of Cloud servers.In this model, the main function of Cloud server is to actas the “control plane” to determine the Fog server neededto be updated with the required content and control datadissemination process, as shown in Fig. 3. Fog servers andpart of Cloud servers are treated as “data plane” to providedata dissemination service.

This data dissemination model has three components,namely asData Structures, Protocol Messages andAlgorithms.


4

Fog

Server

List

Global

Content

List

Mobile

Devices's

Movement

Pattern

Cloud server transmits

this content to the Fog

server through Internet

Mobile devices (carriers)

disseminate this content

to the Fog server based

on DTN technique

Fog

Server

List

Global

Content

List

Mobile

Devices's

Movement

Pattern

Cloud server transmits

this content to the Fog

server through Internet

Mobile devices (carriers)

disseminate this content

to the Fog server based

on DTN technique

Control Plane

Data Plane

Content provider determines the Fog server

needed to be updated and the required content

Cloud Provider determines to use either mobile

devices for DTN data dissemination or cloud

server for traditional data transfer

Fig. 3: Control Plane and Data Plane in This Model

Data structures use tables to store key information which isused to determine the path of data dissemination. Protocolmessages use various tapes of messages to discover contentand mobile devices associated with a Fog server, exchangecontent list, and other tasks to learn and maintain accurate in-formation about the network. Algorithms are used to calculatethe best data dissemination path.

Cloud server in this model needs to have an overall infor-mation and its data structures include the following tables:

• Fog Server List Table: a table to record all Fog servers,which are managed by Cloud server. This table includesFog server’s ID, content ID in each of Fog servers, mobiledevice ID associated with each of Fog servers.

• Global Content List Table: a table to record all publiccontents (not include these private content created by Fogserver owner) in Fog servers or supposed to be in Fogservers. This table includes content ID, the size of eachcontents, Fog server’s ID (for these Fog servers who havethis content), date of update, validation time.

• Mobile Devices’s Movement Pattern Table: a table torecord mobile device’s ID, Fog server ID (whom mo-bile device linked before), linked time, social attribute,geographic movement pattern.

A Fog server needs to have a table to record content ID,the size of this content, mobile device’s ID which linked withthis Fog server, the linked duration of this mobile device. Formobile devices, they need to record the content ID which theycarry on, their movement path, time duration with a Fog serverand its ID.

In order to collect and exchange the above information,

several data messages are used in this model.

• Hello Message between Fog Servers and Cloud Server: anupdate message from a Fog server to its Cloud provider,which includes its content ID and associated mobiledevices ID. This is a triggered message.

• DTN Data Dissemination Request Message: a messagesent from Cloud provider to a Fog server. When Cloudprovider determines there is content need to be updatedfor Fog server A, if Fog server B has this content andits associated mobile device has potential move to Fogserver A, a DTN data dissemination request message issent to Fog server B to ask it to disseminate the contentthrough that mobile device.

• DTN Data Dissemination Accept Message: a messagesent from Fog server to its Cloud provider. Once aFog server received a DTN data dissemination requestmessage and it sends the content to the correspondingmobile device, it sends this DTN data disseminationaccept message to its Cloud provider to confirm that thiscontent has been sent out.

• DTN Data Dissemination Decline Message: a messagesent from Fog server to its Cloud provider. Once aFog server received a DTN data dissemination requestmessage, but it’s associated carrier (mobile device), forsome reason, does not receive the complete content beforethis carrier leave the current Fog server. In this case, theDTN data dissemination decline message is sent back tothe Cloud provider.

• DTN Data Dissemination Acknowledgement Message: amessage sent from Fog server to its Cloud provider. Whena Fog server receives the assigned content from carrier, itsends this acknowledgement message to Cloud providerto confirm that this DTN data dissemination is completed.

Algorithms together with other operations in this hybriddata dissemination model are described in the following sub-section.

B. Hybrid Data Dissemination

Hybrid data dissemination is determined by control plane,as shown in Fig. 3, where Cloud provider in the control planehas the global information to control the data plane. The maindata flow control algorithm conducted by Cloud provider isillustrated in Algorithm 1, where Cloud provider checks itsglobal Fog server and content lists to determine the Fog serverthat needed to be updated, and the required content. It alsochecks which Fog server has this content. If none of Fogserver has this content, Cloud provider sends updated contentto that Fog server directly by using traditional Cloud basedtechniques, such as broadband and cellular networks. Other-wise, the DTN based data dissemination is applied by usingAlgorithm 2 to choose mobile devices, which are connectingwith these selected Fog servers, as carriers to provider DTNbased data dissemination services.

The carriers selection is based on their delivery time anddelivery probability (Algorithm 3). A pre-determined contentdelay threshold,Tdelay, which is an attribute of this contentand can also be treated as content delivery priority, is provided


5

TABLE I: Notations

Notation Definition

Fd A Fog server needed to be updated (thedestination of data dissemination)

C A content needed to be updated and dis-seminated

Fi The Fog serveriCarrieri A content carrier (mobile device)i to

provide DTN based data disseminationservice

Tdelay The maximum affordable delay time of acontent. It is a time period from now toits must updated time, such as a shoppingcenter promotion video must be releasedon every Wednesday

MobListd(i) The ith mobile device (carrier) in themobile list attached with Fog serverd

T imed(i) An average connection period betweenmobile device i and Fog server(d). It isa maximum time window used to up-load/download a content to a Fog server

SizeC The size of contentCSpeedi The wireless transmission speed of mobile

device i

Loci <

Lan, Lon, T >The geographic location vector of mobiledevice i visited at latitudeLan and lon-gitudeLon on timeT

Directioni The expected movement direction of mo-bile devicei

DeliT imei An average content delivery time of Car-rier i to the destined Fog server

ConFrem A contact frequency between the mobiledevice m and its destined Fog server

DeliP robList[n] A list to store all carriers (mobile devices)delivery probability to the destined Fogserver

by the Cloud provider. Only those mobile devices (carriers)with a shorter delivery time compared with the pre-determineddelay time, and a higher delivery probability are selected aspotential DTN based data dissemination carriers.

Once these carriers are chosen, Cloud providers send DTNdata dissemination request message to each of the selectedFog servers to ask them to send the required content to theFog server which is needed to be updated. If the contentis transmitted to the carrier successfully, a DTN data dis-semination accept message is sent back to Cloud providerconfirming this content has been sent out. Otherwise, a DTNdata dissemination decline message is sent. For example, amobile device (carrier) left a Fog server’s coverage area.

When the Fog server, who needs this content, receives thecontent, it sends a DTN data dissemination acknowledgementmessage to Cloud provider to confirm it has received thecontent and this DTN based data dissemination process isfinished. Otherwise, Cloud provider needs to re-select mobilenodes as carriers or directly sends the content using traditionalmethod. Detailed hybrid data dissemination processed areillustrated in Algorithm 1, 2 and 3, and notations used in thesethree algorithm are explained in Table I.

C. Privacy and Reliability of FogRoute

The proposed hybrid data dissemination model in Fogcomputing is promising for resource utilization and network

performance improvement. However, it still has several con-cerns, particularly from privacy and reliability point of views.

How to maintain the privacy of the content? Privacy isa main security concern [17], [18] in the DTN based datadissemination. Selected mobile devices need to store, carryand forward a content from the current Fog server to thedestined Fog server, where the content temporal stored inits storage does not belong to it. In that case, if a mobiledevice steals a glance of this content, it will cause potentialsecurity leakage. For example, a new product is coming tomarket in two days and its video demo and price informationis being distributed among selected fog severs. If one of mobiledevices detects this content and sells this information to thisproduct’s competitor, it will potentially involve a huge lost ofthis product.

To overcome this problem, asymmetric cryptography [19]is deployed in the proposed model. Every Fog server haspublic and private keys, where it stores the private key byitself and upload its public key into Cloud provider, which isthe trustable control plane. When Cloud provider determinescarriers, it passes the destined Fog server’s public key tomobile device’s associated Fog server, where this Fog serveruses the public key to encrypt the content. Without the pairedprivate key, carriers cannot disclose the stored content andcontent’s privacy is protected.

How to make sure mobile devices are willing to carrycontents? A key factor affecting the reliability of DTN baseddata dissemination is whether mobile nodes are willing tostore, carry and forward the assigned content or not. Mobiledevices, particular for those non-power supply devices, havea relative small amount of storage space and limited power.Why they want and how to motive them to participate into theDTN based data dissemination become an important researchissue.

Some incentive DTN forwarding schemes [20] and [21] hadalready been proposed, however the data forwarding amongcarriers when they are travelling to the designated Fog server,is not the main concern in this paper. Instead, we only considerthe “one hop” DTN delivery, where a mobile device downloadsthe required content, stores it into its local storage, carriesto the destined Fog server and forwards this content to it.A light incentive scheme is deployed in this model, whichlikes a “loyal membership”. Mobile devices who are willingto deliver content and have a high delivery success ratio willbe rewarded a high membership points. Membership pointscan be used to subscribe those pay-video services or redeem ahigher discount voucher from Fog servers or Cloud provider.

IV. SIMULATION

Simulation of the proposed hybrid data dissemination ispresented in this section to show the performance of FogRoute.As data dissemination through Cloud infrastructure is a verymature field, the main focus in our simulation is the per-formance of DTN based off loading. The first sub-sectionillustrates the data set we are going to use and the simulationset-up. The second sub-section analyzes the simulation results.


6

Algorithm 1 Data Flow Control Algorithm

Step 1:Cloud provider compares its “Global Content List” table with “Fog Server List” table to determine which Fog serveris needed to be updated. In this case, Fog severFd is determined and contentC needs to be updated.if There is no other Fog server having this contentthen

Cloud provider sends this content to the Fog server directly, which is the same as traditional Cloud service and thisdissemination process is finished.

elseMove to the next step

end ifStep 2: Cloud provider determines a list of Fog servers,< Fc1, Fc2, Fc3, · · · >, which have this content.Step 3: Algorithm 2 is used to selectn most suitable carriers,Carriern, to provide this DTN dissemination service.if n greater than0 then

Move to the next stepelse

Cloud provider sends this content to the Fog server directly, which is the same as traditional Cloud service and thisdissemination process is finished.

end ifStep 4: Cloud provider sends “DTN Data Dissemination Request” message to each of selected Fog servers to ask themsend the contentC to Fd by using the carrier (mobile device) determined inStep 3.Step 5: Once a Fog server receives “DTN Data Dissemination Request” message, it sends the contentC along with thedestination,Fd, to the selected carrier.if ContentC is transferred to the selected carrier completelythen

This Fog server sends the “DTN Data Dissemination Accept” message to its Cloud provider and move to the next step.else

This Fog server send the “DTN Data Dissemination Decline” message to its Cloud provider. When Cloud provider receivesthis message, it repeatsStep 3 to get the “n+1” Fog server, if it has, and continue fromStep 4.

end ifStep 6:OnceFd receives the contentC, it sends “DTN Data Dissemination Acknowledgement” message to Cloud provider.Step 7:if Cloud provider receives the “DTN Data Dissemination Acknowledgement” message within a pre-defined period,Tdelay,in Algorithm 2 then

It updates “Fog Server List” and “Global Content List” tables, and this data dissemination is finishedelse

It repeats from theStep 1end if

A. Simulation Set-up

To simulate the DTN based data dissemination and particu-lar for their mobile carriers’ selection, we use the real taxi tracefile and the public transport data, such as bus, from the citycenter of Rome, Italy generated by Raul et. al [22]. The dataset collects370 taxi cabs’ movement trace in the city of Romefor a period of6 months. Each of taxis is equipped with anAndroid OS tablet running an app to update its current positionto a server in every7 seconds with its latitude and longitude,and this trace file only records those traces occurred in thecenter of Rome within a8kmx8km area with the coordinatespairs as(41.856, 12.442) and (41.928, 12.5387). As the cityof Rome is characterized by very thin and congested roads,the average speed of taxis is slow and only12.29km/h.

For the simulation purpose, the one month most stable tracecollected from February 1st, 2014 to March 2th, 2014 is used.The Fig. 4 demonstrates the overall reported positions of these370 taxis in one day period, which is indexed by its latitudeand longitude in the8kmx8km area. As some locations arereported more than6000 times, while there are still some

locations are reported only0 or 1 time, to make this figuremore readable, we capped the daily reported frequency as300times. According to this figure, majority of place in this areahave been visited by at least one taxi, as long as they are carreachable. If we further use the investigation in [22] where therealistic wireless coverage of the vehicle-mounted Android OStablet is50m by using2.4GHz and5GHz communications,the wireless coverage area of each taxi is becoming muchlarger.

In addition of these non-scheduled taxi traces, we re-trieved the schedule of bus traces in the city of Rome fromATAC (Tramways Company and Coach of the Municipality ofRome) [23], which has all public transport data of Rome, suchas bus route, travel frequency and travel time between two busstops. Bus routes coverage of the proposed simulated area isabstracted in Fig. 5. Wireless coverage of these scheduled busservices cover the most popular tourism area and cafe shopsin the city area, where Fog server could be deployed.50 Fogservers are deployed in the simulation areas to provide datadissemination services.


7

Algorithm 2 DTN Data Dissemination Carrier Selection Algorithm

Step 1: Cloud provider determines the affordable delay time,Tdelay, of this content.Tdelay Cloud be treated as the priorityof this content.Step 2: Cloud provider checks its “Fog Server List” and “Mobile Devices’s Movement Pattern” tables to find the list ofmobile users accessedFd before,MobListd(i), and their average connection time withFd, T imed(i), wherei is the ID ofconnected mobile device.Step 3: These mobile devices with a short connection time are filtered out, as they are not able to upload the content to theFog server:for each of mobile device inMobListd(i) do

ifSizeCSpeedi

> Timed(i)

thenThis mobile device is filtered out fromMobListd(i) and a new listMobList′d(r) is formed, wherer is the number ofmobile device meeting the above condition

end ifend forStep 4: MobList′d(r) is further classified into two categories,scheduled andnon-scheduled visit lists. Scheduled visit liststores these mobile devices which are pre-determined to visit a particular Fog server, such as airport shuttle bus. The restof filtered mobile devices are classified into the non-scheduled visit list.Step 5: For scheduled mobile devices,Si, as long as its delivery time, which is the time from now to itsnext scheduled visit time, is within theTdelay, it is added into the DTN data dissemination carrier list,<CarrierS1, CarrierS2, · · · , CarrierSx >, wherex the total number of carriers selected to add into the carrier list.Step 6: Non-scheduled mobile device list,NSi, is re-ordered by mobile devices’ delivery probability toFd based onAlgorithm 3. Cloud provider select the topy mobile devices according their delivery probabilities to add them into the DTNdissemination carrier list, where the number ofy is the largest number to satisfy the following condition:

∑x

i=1DeliT imeSi +

∑y

i=1DeliT imeNSi

x+ y< Tdelay

End: Now x+ y mobile devices are selected as carriers to provide DTN data dissemination service.

Fig. 4: Frequency of Taxi Reported Locations in the Center ofRome

To conduct this simulation, we implement a Java baseddiscrete event driven simulator. This simulator simulated taxisand buses’ movements with their times and locations.50Fog servers are deployed in the simulation area with randomlatitudes and longitudes. Contents used in this simulator wererandomly generated and their sizes are in the range of1MBs

Fig. 5: Abstracted Bus Routes in Rome with bounds’ coordi-nates pairs as(41.856, 12.442) and (41.928, 12.5387)

to 6GBs, and they are distributed among in different Fogservers. All mobile devices’ movements, content distribution,Fog server associated devices etc. are stored in central tosimulate the Cloud environment.

Two main experiments are conducted to test the perfor-


8

Algorithm 3 Mobile Device Delivery Probability

Step 1: For each of mobile devices,m, Cloud provider collects its contact frequency withFd, ConFrem, geographiclocations and visit times of the three most recently visited Fog servers,Locm < Lan,Lon, T >.Step 2: Based on the three most recent visited history,Locm < Lan,Lon, T >, and real distance from the map, averagemovement speed and direction of mobile devicem could be generated asSpeedm andDirectionm.Step 3: The expected delivery time from mobile devicem to Fog serverFd, DeliT imem, is calculated by usingSpeedm,Directionm and the geographic distance between both them.if DeliT imem > Tdelay then

The delivery probability of this mobile device,DeliProbm, is marked as0 and this algorithm is finishedelse

Move to next stepend ifStep 4: Assume there aren suitable mobile devices left in this step. The overall delivery probability of mobile devicem iscalculated as:

DeliProbm =ConFrem∑n

i=1ConFrei

× (1−DeliT imem∑n

i=1DeliT imei

)

and each of them is added into the delivery probability list,DeliProbList[n].Step 5: SortDeliProbList[i] in ascending orderSetu = 1, j = nwhile u ≤ n do

while j > u doif DeliProbList[j − 1] > DeliProbList[j] then

swap(DeliProbList[j − 1], DeliProbList[j])end ifj −−

end whileu++

end whileThis delivery probability list is ready to be used for Algorithm 2.

mance of the proposed model. The first simulation is to testthe affect of expected delay to the final delivery success ratio.The second experiment is to test the convergence time of datadissemination for each of Fog servers. All experiments aresimulated10 times and their average results are recorded.

B. Simulation Result

Delivery success ratio is the main concern and key indicatorwhen designing and selecting a data dissemination model. It isparticularly important for this hybrid model in Fog computing,as the main target in this model is to disseminate content intothe selected Fog server within the expected delay time. In thatcase, the content provider or Cloud administer needs to finda suitable expected delay from both economic and reliabilitypoint of views.

Fig. 6 shows the average delivery success ratios withdifferent expected delays. In this experiment, the content to bedisseminated and the Fog server whose content needed to beupdated are randomly chosen. This result demonstrates that theexpected delay has a great impact to the delivery success ratio.When the expected delay is less than2hrs, the average deliversuccess ratio is less than50%. When the expected delay timeis increased to6hrs, the average deliver success ratio reachesto 80%. When the expected delay time is larger than8hrs,

Expected Delay (in mins)0 100 200 300 400 500 600 700 800

Del

iver

y S

ucce

ss R

atio

0

10

20

30

40

50

60

70

80

90

100

Fig. 6: Performance of Delivery Success Ratio with DifferentExpected Delay

the delivery success ratio is more than90%. After 10hrs, thedelivery success ratio is converged at95%.

The above results give a clear picture of how to use theproposed hybrid data dissemination model. For example, whenthe content dissemination is not that urgent, say a half of


9

Fog Server ID0 5 10 15 20 25 30 35 40 45 50

Con

verg

ence

Tim

e (in

min

s)

0

100

200

300

400

500

600

700

Fig. 7: Performance of Convergence Time for Each of FogServers

day delay is affordable, the DTN based data disseminationis the best choice from both economic and reliability point ofview. If the content needed to be disseminated to other Fogservers in one or two hours, such as important news, DTNbased data dissemination could be used, but a Cloud baseddata transmission channel should be ready as a backup.

Convergence time is another key factor to reflect howefficient of a data dissemination model, which measures thetime spent on a Fog server to receive all contents. In thisexperiment, we randomly select20 contents from the list andset up their destinations to each of Fog servers. We recordthe time for each of Fog servers to receive a content and theconvergence time is the total time of a Fog server to receivethese20 contents. To make this content dissemination moregeneral, size of contents are vary from1MBs to 6GBs andthere are50 Fog servers all together.

Fig. 7 shows the average convergence time for each of Fogservers, where7 Fog servers could complete the data dissem-ination in just10mins! In summary, there are15 Fog serverscould reach the convergence in1hr and27 Fog servers reachthe convergence in2hrs, which means the data disseminationto more than half of total Fog servers with20 contents couldbe completed within the first2hrs. In addition, there are46Fog servers all together could reach the convergence within11hrs, while4 Fog servers cannot be converged within12hrs.

The above results demonstrate the proposed model is verysuitable for a large number of content dissemination based onDTN technique. It can be seen that92% of overall contentscould be successfully disseminated to each of Fog serverswithin a day, and there are only4% of Fog servers cannotsuccessfully update their content list, where a higher costCloud based data dissemination is therefore required to bedeployed.

Compared with the traditional model, another advantageof the proposed model is on the public bandwidth saving.With the advances in modern Cloud network technique, thecost per unit bandwidth has dramatically decreased, which,

Content ID0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Pub

lic B

andw

idth

Sav

ing

Rat

io

0

20

40

60

80

1001hr Expected Delay3hrs Expected Delay6hrs Expected Delay

Fig. 8: Performance of Public Bandwidth Saving

however, is accompanied by the surge increase data volumeto be transmitted, such as high resolution videos. e.g. for a 1TBbandwidth on Microsoft Azure Cloud service, it takes around100 dolors.In this experiment, the ratios of public bandwidthsaving with different expected delay are demonstrated.

Fig. 8 demonstrates the performance of public bandwidthsaving for three different metrics, namely 1hr, 3hrs and 6hrsexpected delays, where the expected delay is used to identifythe affordable delay for a certain content. 20 contents are ran-domly selected and their bandwidth saving ratios are displayedin bars. It can be seen that when the expected delay is 1hr, theaverage bandwidth saving ratio is30%, and with the expecteddelay increased to 3hrs, the average bandwidth saving ratio is55%. Once the expected delay is increased to 6hrs, the averagebandwidth saving ratio reaches to85%.

The simulation results demonstrate that the proposed modelcan dramatically decrease the public bandwidth usage, whilemaintain a high data dissemination standard. It shows that85%of public bandwidth can be saved in the simulated scenario,when the delay of data dissemination is up to 6hrs. For thosehigh volume and delay-tolerable contents, such as video andmovie, 6hrs is a reasonable and affordable delay time.

Is summary, these experiment results show the proposed hy-brid data dissemination model not only saves public bandwidthand utilizes network resource, but also has a higher deliverysuccess ratio with a lower end to end delay. It proves theFogRoute is a workable, efficient and reliable model in Fogcomputing paradigm.

V. CONCLUSION AND FUTURE WORK

In this paper, we have proposed a hybrid data disseminationmodel in Fog computing by utilizing both delay tolerantnetwork and Cloud techniques. These contents with highervolume and lower timeliness, such as high definition video,could be disseminated by delay tolerant technique. While thesecontents with lower volume and higher timeliness, such asemergency news, could be disseminated with current Cloud


10

technique. The proposed model is central controlled by con-tent provider or Cloud administer and all Fog servers aredistributed deployed in local areas. Delay tolerant networkbased data dissemination is based on human (mobile devices)and vehicles. Experiment results demonstrated the proposedmodel has a higher delivery success ratio and lower end to enddelay, which proved it is not only workable, but also reliableand efficient in term of data dissemination in Fog computing.

For the future work, we intend to extend the current proposalto a more generalized scenario as follows. The currently carrierforwarding algorithm in delay tolerant network is based onone hop delivery only, where carriers download content fromcontent provider (source Fog server), carry to the destinationFog server and upload the content to it. In that case, if amobile device has limited connect with the destination Fogserver, it cannot be selected as carrier. In the future work, theforwarding algorithm could be designed as relay-exchange-forwarding, where mobile devices could exchange/multi-castcontents among themselves when they are moving to the finaldestination.

REFERENCES

[1] F. Bonomi, R. Milito, J. Zhu, and S. Addepalli, “Fog computing andits role in the internet of things,” inProceedings of the First Edition ofthe MCC Workshop on Mobile Cloud Computing, ser. MCC ’12. NewYork, NY, USA: ACM, 2012, pp. 13–16.

[2] Cisco, “Cisco delivers vision of fog computing to accelerate value frombillions of connected devices,”Cisco Press Release, Jan 2014.

[3] T. H. Luan, L. Gao, Z. Li, Y. Xiang, and L. Sun, “Fog Computing:Focusing on Mobile Users at the Edge,”ArXiv e-prints, Feb. 2015.

[4] C. W. Paper, “Cisco visual networking index: Global mobile data trafficforecast update, 2014?019,” Feb 2015.

[5] I. Stojmenovic and S. Wen, “The fog computing paradigm: Scenariosand security issues,” inComputer Science and Information Systems(FedCSIS), 2014 Federated Conference on, Sept 2014, pp. 1–8.

[6] R. Deng, R. Lu, C. Lai, T. H. Luan, and H. Liang, “Optimal workloadallocation in fog-cloud computing towards balanced delay and powerconsumption,”IEEE Internet of Things Journal, May 2016.

[7] Z. Xu, W. Liang, W. Xu, M. Jia, and S. Guo, “Efficient algorithms forcapacitated cloudlet placements,”IEEE Transactions on Parallel andDistributed Systems, vol. 27, no. 10, pp. 2866 – 2880, October 2015.

[8] Y. Zheng, L. Capra, O. Wolfson, and H. Yang, “Urban computing:Concepts, methodologies, and applications,”ACM Trans. Intell. Syst.Technol., vol. 5, no. 3, pp. 38:1–38:55, Sep. 2014.

[9] L. Gao, S. Yu, T. H. Luan, and W. Zhou,Delay Tolerant Networks andTheir Applications. Springer Publishing Company, Incorporated, 2015.

[10] Y. Cao and Z. Sun, “Routing in delay/disruption tolerant networks: Ataxonomy, survey and challenges,”Communications Surveys Tutorials,IEEE, vol. 15, no. 2, pp. 654–677, Second 2013.

[11] A. Pentland, R. Fletcher, and A. Hasson, “DakNet: rethinking connec-tivity in developing nations,”Computer, vol. 37, no. 1, pp. 78–83, 2004.

[12] L. Gao, M. Li, A. Bonti, W. Zhou, and S. Yu, “Multidimensionalrouting protocol in human-associated delay-tolerant networks,”MobileComputing, IEEE Transactions on, vol. 12, no. 11, pp. 2132–2144, Nov2013.

[13] M. Li, R. Na, Q. Qian, H. Zhu, X. Liang, and L. Yu, “Spfm: Scalableand privacy-preserving friend matching in mobile cloud,”IEEE Internetof Things Journal, June 2016.

[14] Z. Su, Q. Xu, H. Zhu, and Y. Wang, “A novel design for content deliveryover software defined mobile social networks,”IEEE Network, vol. 29,no. 4, pp. 62–67, June 2015.

[15] T. H. Luan, L. Cai, J. Chen, X. Shen, and F. Bai, “Engineering adistributed infrastructure for large-scale cost-effective content dissem-ination over urban vehicular networks,”Vehicular Technology, IEEETransactions on, vol. 63, no. 3, pp. 1419–1435, March 2014.

[16] L. Gao, M. Li, A. Bonti, W. Zhou, and S. Yu, “M-Dimension: Multi-characteristics based routing protocol in human associated delay-tolerantnetworks with improved performance over one dimensional classicmodels,”Journal of Network and Computer Applications, vol. 35, no. 4,

pp. 1285 – 1296, 2012, intelligent Algorithms for Data-Centric SensorNetworks.

[17] H. Zhu, S. Du, Z. Gao, M. Dong, and Z. Cao, “A probabilisticmisbehavior detection scheme toward efficient trust establishment indelay-tolerant networks,”IEEE Transactions on Parallel and DistributedSystems, vol. 25, no. 1, pp. 22–32, 2014.

[18] Z. Tang, A. Liu, Z. Li, Y. june Choi, H. Sekiya, and J. Li, “A trust-basedmodel for security cooperating in vehicular cloud computing,”MobileInformation Systems, vol. 2016, 2016.

[19] A. Targhetta, D. Owen, and P. Gratz, “The design space of ultra-lowenergy asymmetric cryptography,” inPerformance Analysis of Systemsand Software (ISPASS), 2014 IEEE International Symposium on, March2014, pp. 55–65.

[20] X. Zhuo, W. Gao, G. Cao, and S. Hua, “An incentive framework forcellular traffic offloading,”Mobile Computing, IEEE Transactions on,vol. 13, no. 3, pp. 541–555, March 2014.

[21] R. Lu, X. Lin, H. Zhu, X. Shen, and B. Preiss, “Pi: A practical incentiveprotocol for delay tolerant networks,”Wireless Communications, IEEETransactions on, vol. 9, no. 4, pp. 1483–1493, April 2010.

[22] R. Amici, M. Bonola, L. Bracciale, A. Rabuffi, P. Loreti, and G. Bianchi,“Performance assessment of an epidemic protocol in{VANET} usingreal traces,”Procedia Computer Science, vol. 40, pp. 92 – 99, 2014,fourth International Conference on Selected Topics in Mobile WirelessNetworking (MoWNet?014).

[23] “Atac, company for rail and road transport of the city of rome,”http://viaggiacon.atac.roma.it/index.html?language=eng, accessed: 2015-02-26.

Longxiang Gao (S’09, M’11) received his PhDin Computer Science from Deakin University, Aus-tralia. He is currently a Lecturer in Computer Net-works at School of Information Technology, DeakinUniversity. Before he joined Deakin University, hewas a post-doctoral research fellow at IBM ResearchAustralia. His research interests include mobile so-cial networks, delay tolerant networks and Fog com-puting. His research works have been published inmany international journals and conferences, suchas IEEE Transactions on Mobile Computing. He

received 2012 Chinese Government Award for Outstanding Students Abroad(Ranked No.1 in Victoria and Tasmania consular districts). Dr. Gao is active inIEEE Communication Society. He has served as the TPC co-chair, publicty co-chair, organization chair and TPC member for many international conferences.He is a member of IEEE and ACM.

Tom H. Luan (M’14) is a Lecturer in the Schoolof Information Technology at the Deakin Universityin Australia. He received his Ph.D. degree from theDepartment of Electrical and Computer Engineeringat the University of Waterloo in 2012. From March2013 to August 2013, he was a visiting researchscientist in the Institute of Information Engineer-ing, Chinese Academy of Sciences. Since December2013, Dr. Luan has joined the Deakin University,where he is a Lecturer on the Mobile and App. Hisresearch mainly focuses on the areas of mobile cloud

computing, mobile app and service development, vehicular networking andwireless content distribution. Dr. Luan has published about thirty technicalpapers in journals and conference proceedings. He served as a TPC memberfor IEEE Globecom, ICC, PIMRC and the technical reviewer for multipleIEEE Transactions including TMC, TPDS, TVT, TWC and ITS. He isa committee member of the Interest Group on Wireless Technology forMultimedia Communications of IEEE ComSoc Multimedia CommunicationsTechnical Committee (MMTC). He is a member of IEEE.


11

Shui Yu (SM’12) currently a Senior Lecturer of theSchool of Information Technology, Deakin Univer-sity, Australia. Dr Yu current research interests in-clude Big Data, network security, privacy and foren-sics, networking theory, and mathematical modeling.He targets on narrowing the gap between theoryand applications using mathematical tools. He haspublished more than 150 fully reviewed papers,including top journal papers and top conferencepapers, such as IEEE Transactions of Parallel andDistributed Systems, IEEE Transactions on Com-

puters, IEEE Transactions on Information Forensics and Security, IEEETransaction on System, Man, and Cybernetics, Part B, IEEE Transactions onVehicular Technology, IEEE Transactions on Mobile Computing, and IEEEINFOCOM. Dr Yu had one monograph (Distributed Denial of Service Attackand Defence, Springer), edited a book titled Networking for Big Data (CRC,2015).

Dr Yu is active in research services in various roles. He serves theeditorial board of IEEE Transactions of Parallel and Distributed Systems,IEEE Communications Survey and Tutorial, IEEE Access, and three otherInternational journals. He served as a leading guest editor for a numberof special issues, including Networking for Big Data on IEEE Networkin 2014, Big Data for Networking on IEEE Network in 2015. He is alsoinvolved in organizing international conferences, such as TPC co-chair ofIEEE Big Data Computing Service and Application in 2015, founder ofInternational Workshop on Security and Privacy in Big Data (usually withIEEE INFOCOM), symposium co-chair for CISS of IEEE ICC 2014, IEEEICNC 2014, publication chair for IEEE Globecom 2015 and IEEE INFOCOM2016, TPC member for INFOCOM 2012-15. He is a senior member of IEEE.

Wanlei Zhou (SM’09) received the B.Eng andM.Eng degrees from Harbin Institute of Technology,Harbin, China in 1982 and 1984, respectively, andthe PhD degree from The Australian National Uni-versity, Canberra, Australia, in 1991, all in Com-puter Science and Engineering. He also received aDSc degree from Deakin University in 2002. Heis currently the Alfred Deakin Professor and Chairof Information Technology, School of InformationTechnology, Deakin University. Professor Zhou hasbeen the Head of School of Information Technology

twice (Jan 2002-Apr 2006 and Jan 2009-Jan 2015) and Associate Dean ofFaculty of Science and Technology in Deakin University (May 2006-Dec2008). Before joining Deakin University, Professor Zhou served as a lecturerin University of Electronic Science and Technology of China, a systemprogrammer in HP at Massachusetts, USA; a lecturer in Monash University,Melbourne, Australia; and a lecturer in National University of Singapore,Singapore. His research interests include distributed systems, network security,bioinformatics, and e-learning. Professor Zhou has published more than 300papers in refereed international journals and refereed international conferencesproceedings. He has also chaired many international conferences and has beeninvited to deliver keynote address in many international conferences. ProfessorZhou is a Senior Member of the IEEE.

Bo Liu (S’09, M’10) received the B.S. degree fromNanjing University of Posts and Telecommunica-tions, Nanjing, China, in 2004, and the M.S. andPh.D. degrees, in 2007 and 2010, respectively, fromShanghai Jiao Tong University, Shanghai, China,all in electrical engineering. Since 2010, he hasbeen an Assistant Researcher and then AssociateResearcher at the Electrical Engineering Departmentof Shanghai Jiao Tong University. He is currentlya Postdoc Research Fellow in Deakin University,Australia, since November 2014.

Dr. Liu is active in IEEE Communications Society and IEEE BroadcastingSociety. He has served as the TPC member and session chair of many IEEEconferences and the reviewer of IEEE transactions and journals.

Documents

FogRoute: DTN-based Data Dissemination Model in Fog Computing · FogRoute: DTN-based Data Dissemination Model in Fog Computing Longxiang Gao Member IEEE, ... Fog servers, an important