Sporadic Cloud Computing over a Virtualization Layer

A New Paradigm to Support Mobile Multi-hop Ad-hoc Networks

Esteban F. Ord´o˜nez-Morales

, Yolanda Blanco-Fern´andez

and Mart´ın L´opez-Nores

Grupo de Investigaci´on e Innovaci´on en Ingenier´ıas, Universidad Polit´ecnica Salesiana, Cuenca, Ecuador

AtlantTIC Research Center, Department of Telematics Engineering, University of Vigo, Vigo, Spain

In our doctoral proposal we deploy Sporadic Ad-

hoc Networks (SANs) over the devices of a group

of always-on users who happen to meet in a place.

The goal is to develop tailor-made services that ex-

ploit the possible similarities among the preferences

of the users and the technological capabilities of their

terminals to establish direct and hop-by-hop ad-hoc

communications. In order to overcome the intrin-

sic limitations of mobile devices, we explore the new

concept of Sporadic Cloud Computing (SCC) that is

aimed at providing each terminal with additional re-

sources by exploiting the (computational, network-

ing, storing...) capabilities of the rest of devices con-

nected to the SAN. In order to abstract the complex-

ity stemmed from the mobility scenarios, SCC works

with an enhanced Virtualization Layer that deals with

a few static virtual nodes instead of a higher num-

ber of mobile real nodes. This allows to turn our

SANs into reliable and stable communication envi-

ronments to promote interactions among potentially

like-minded strangers in a great diversity of mobility

scenarios, involving both pedestrians and cars in ve-

hicular environments.

1 RESEARCH PROBLEM

After the irruption of the Web 2.0 and the smartphone

revolution, most of the on-moveusers carry with them

handheld devices for actively interacting daily with

their friends/followers/followees/... in the context of

the virtual world of the Internet. Sociologist have al-

ready advised about the negative effects derived from

some of these behaviours, which might lead the users

to immersing themselves in a virtual communication

burble with their contacts, by giving up interacting

face-to-face with nearby individuals (Kuss and Grif-

ﬁths, 2011). In the same line, other experts have

analyzed the consequences of the so-called FOMO

(Fear Of Missing Out) effect which denotes the fear

of users of losing events, news and situations that hap-

pen in their social networks, thus not taking an eye

off their devices (Przybylski et al., 2013). In order

to ﬁght these situations, we are working in the devel-

opment of the SPORANGIUM (SPORAdic networks

in the Next-GenerationInformation services for Users

on the Move) platform which deployes Sporadic Ad-

hoc Networks (SANs) among always-on users who

happen to be in a place, by establishing multi-hop

ad-hoc connections over their respective mobile ter-

minals. The goal is to promote more direct interac-

tions among strangers who happen to meet in spaces

like cinemas, stadiums, museums, concert halls, etc.,

who might have potentially common interests (cin-

ema, sport, art, music, etc.) that would be convenient

to explore.

For that purpose, SPORANGIUM must orches-

trate activities and tailor-made services that bring to-

gether the particular context of the users, their po-

tentially common preferences and the capabilities

of their devices for establishing ad-hoc connections.

However, these services far exceed traditional mobile

devices capabilities, which suffer from computational

limitations, as well as battery restrictions and pro-

cessing time. To face this situation, the new Mobile

Cloud Computing (MCC) paradigm has arisen that

takes inspiration from the well-known Cloud Com-

puting (CC), which is based on delivering comput-

ing as a service whereby shared resources, software

and information are provided as a utility over a net-

work (typically the Internet). In MCC the goal is to

enable to process a large amount of data on demand

anytime from anywhere, so that mobile devices con-

nect to the Internet to use an environment that inte-

grates diverse platforms and technologies (Dinh et al.,

2013). Speciﬁcally, MCC promotes to move the com-

puting power and data storage away from mobile de-

vices and into the cloud, bringing multiple service

models (IaaS, PaaS, SaaS...) and mobile computing

to a wide range of on-move always-on users.

Recently, the MCC paradigm has been exported to

vehicular communication environments where some

researchers have proposed the so-called Vehicular

Cloud Computing (VCC). The goal here is to exploit

F. Ordóñez-Morales E., Blanco-Fernández Y. and López-Nores M..

Sporadic Cloud Computing over a Virtualization Layer - A New Paradigm to Support Mobile Multi-hop Ad-hoc Networks.

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

both the physical data center units that are in charge

of performing the data computation and storage (like

in MCC) and the on-board resources of the own ve-

hicles (Olariu et al., 2013). In our proposal of PhD

work, we want to extend the VCC paradigm to sup-

port mobility scenarios beyond the vehicular environ-

ment, by involving both pedestrians and vehicles on

the road. Speciﬁcally, we explore a new paradigm –

named Sporadic Cloud Computing (SCC)– aimed to

allow the users’ devices to exploit both the (comput-

ing, storing, networking, sensing...) resources avail-

able in the rest of terminals connected to the SANs,

and those provided from external data centers. In

the deployment of our SCC paradigm in diverse mo-

bile communicationenvironments,one of the main re-

search challenges has to do with the high mobility of

the nodes connected to the ad-hoc network (e.g. cars

in a vehicular ad-hoc network), and therefore, with its

frequenlty changing topology. This causes that com-

munication fails often as these nodes move fast and

are out of the range of the ad-hoc network, which also

hampers the routing tasks when forwarding informa-

tion (Gerla, 2012).

2 OUTLINE OF OBJECTIVES

Our doctoral proposal is aimed at designing, develop-

ing and validating the mechanisms necessary to:

1. turn our SANs networks into reliable and stable

communication environments with good perfor-

mance in terms of overhead, packet delivery ratios

and scalability, covering vehicular, pedestrian and

mixed environments, and

2. deploy enhanced “X”aaS service models (e.g.

CaaS, NaaS, STaaS, SEaaS...) through our SCC

paradigm, so that the devices connected to the

SAN can collaborate and share their respective

Computing, Networking, SToring and SEnsing

capabilities in the deployment of advanced com-

munication services.

As introduced before, the main research chal-

lenges derived from both objectives has to do with

(i) the frequent topology changes happened in certain

communication environments (e.g. in a vehicular ad-

hoc network due to the fast movements of the cars),

and (ii) the fact that the capabilities available in the

SANs vary on the time, as the location of the users’

devices change.

To deal with the high mobility of the nodes con-

nected to our SANs, our proposal takes advantage of

the improvements proposed in the realm of Mobile

Ad-hoc Networks (MANETs), which have been en-

visaged to face the problems derivedfrom (i) the wire-

less transmission mediums, (ii) the high variability

of the network topology due to unpredictable nodes’

movements, and (iii) the existence of severe restric-

tions in terms of processing capabilities, memory and

battery consumption. In particular, we start from the

work presented in (Dolev et al., 2004) where the au-

thors described a virtualization layer named VNLayer

(Virtual Node Layer). Speciﬁcally, the VNLayer is a

cluster-based approach where the mobile nodes col-

laboratively create an infrastructure of static virtual

nodes to ease the routing problem and the mainte-

nance of persistent state information in the area cov-

ered by an ad-hoc wireless network of mobile devices

(as our SANs), notwithstanding the mobility of the

(real) physical nodes. Actually, the VNLayer resides

between the link layer and the Internet Layer, so that

the virtual nodes can be addressed as if they were

static server devices. This helps to mask the uncer-

tainty that arises from the MANETs’ varying topol-

ogy and from the fact that the physical devices can

fail unpredictably. Consequently, it is easier for de-

velopers to work at the nodes’ upper layers, since they

can deploy applications on mobile devices and vir-

tual servers with greater ease and efﬁciency. Besides,

virtualisation creates a level of hierarchy in the oth-

erwise ﬂat MANETs, which brings in opportunities

to re-design MANET protocols to operate more efﬁ-

ciently and reliably.

Since the virtualization layer by Dolev et al.

has been developed to handle communications in

MANETs, the ﬁrst objective of our doctoral proposal

consists of extending and adapting the working of

the VNLayer to the restrictions and peculiarites of

more demanding mobility scenarios, including, for

instance, communication environments where pedes-

trians and occupants of vehicles are involved. To this

aim, we need to envisage reﬁnements in the VNLayer

(resulting in our VNLayer+) to take into account, for

instance, the comparatively faster movements of ve-

hicles, the freedom of movement of the pedestrians,

as well as the fact that these nodes are not subject

to the strict energy, space and computing capabilities

restrictions of MANETs. These restrictions must be

considered to turn our SANs into reliable communi-

cation environments, covering a wide diversity of ap-

plication scenarios beyond the generic MANETs ex-

plored in Dolev et al.’s approach.

Our SCC paradigm must develop transport-layer

coordination mechanisms among the devices that

are connected to the SANs in order to enable an

efﬁcient sharing and allocation of their available

resources by working over the virtualization layer,

SporadicCloudComputingoveraVirtualizationLayer-ANewParadigmtoSupportMobileMulti-hopAd-hocNetworks

which is the second objetive of the doctoral pro-

posal. This fact causes that, differently from the

traditional approaches envisaged in CC, MCC and

VCC, our “X”aaS service models need to deal with

the (static) virtual nodes of the VNLayer+, which

are emulated/supported by the devices of the users

on the move. Speciﬁcally, while the SAN is estab-

lished among the users’ terminals, the messages to

request resources from the ad-hoc network (or to ad-

vertise resources that are left to other devices’ dis-

posal) are managed by virtual nodes. The tandem

SCC-VNLayer+ contributes to (i) ﬁght/alleviate the

communication errors and data loss noticeable in mo-

bile ad-hoc networks (Dinh et al., 2013), and (ii) to

orchestrate advanced applications to improve the ex-

perience of the users, by taking advantage of the reli-

able data exchange over the SAN (thanks to the VN-

Layer+) and the availability of additional resources in

each terminal (thanks to the service models of SCC).

The possible applications to be deployed in the

realm of our SANs cover a wide spectrum, rang-

ing from the orchestation of activities bounded to an

event where a group of like-minded users happen to

meet (e.g. in a museum, theater or stadium), to the

provision of both improved applications for interve-

hicular communication (e.g. optimization of trafﬁc

ﬂows, chats among drivers, proactive organization of

ride-sharing opportunities or selective distribution of

personalized advertising in nearby places), and re-

ﬁnements in the context of the smart cities through

the planiﬁcation of people mobility and urban games,

among others.

3 STATE OF THE ART

In this section, we review related works in the two

main research ﬁelds of our doctoral proposal: the use

of virtualization in MANETs and the exploitation of

resource sharing among devices in mobile communi-

cation environments.

3.1 Virtualization in Mobile Ad-hoc

Networks

The VNLayer was presented in (Dolev et al., 2004) as

a set of procedures to turn ad-hoc networks of mobile

devices into more predictable environments for com-

munications. The main idea is to engage the mobile

physical nodes (PNs) in collaboration to emulate vir-

tual nodes (VNs) that remain in known grid locations,

as shown in Figure 1 (where back circles and white

squares denote PNs and VNs, respectively).

The VNLayer divides the geographical area of an

ad-hoc network into square regions, whose size is

chosen so that every PN in a region can reliably send

and receive data from every other physical node in

that region and neighboring ones. Each VN (one per

region) is emulated by the PNs located in the corre-

sponding region, so that when all the physical nodes

leave this region, the virtual node stops to work. In

each region, one PN is chosen as the leader in the re-

gion and becomes the primary responsible for packet

reception, buffering and forwarding. Meanwhile, a

subset of non-leader nodes are designated as backups

to maintain information consistent with the leader’s

version (speciﬁcally, replicas of the virtualization-

related state information and the routing tables tack-

led by the routing protocols working on the virtualiza-

tion layer). This way, the VNs can maintain persistent

state and be fault tolerant even when individual PNs

fail or leave the region.

Figure 1: Static virtual nodes (white squares) overlaying the

mobile nodes of a MANET (black circles).

An exhaustive analysis of the VNLayer allows to

detect certain sources of inefﬁciency in its function-

ing, which are mainly related to:

• The Procedure used to Identify New Backup

Nodes: The approach adopted by Dolev et al.

in order to designate backup nodes among non-

leaders is a probabilistic one and it is driven by

a Coin Tosser Function, according to which the

greater the number of nodes in the region, the

lower the probability that a PN will choose as

backup. This avoids having many backups in

dense MANETs, and thus reduces the overhead

due to state synchronisations, i.e. to the exchange

of messages aimed at ensuring that the replicas of

the state information from the upper layers (e.g.

routing tables) kept in the backup nodes are con-

CLOSER2015-DoctoralConsortium

sistent with the leader’s version.

• The Procedure adopted in the Leader Election:

This process suffers from two main problems: (i)

it involves a great number of messages to be ex-

changed –which contributes to increase the dura-

tion of the virtual nodes’ downtimes–, and (ii) the

selection process does not prioritize the backup

nodes that have an updated version of the out-

going leader node (which would be the best can-

didates to become a new leader). In particular,

this procedure could designate as new leader ei-

ther a node that was not acting as a backup or

a non-synchronised backup node, even in cases

that there were synchronised backups in the re-

gion. This causes that the state information from

the upper layers would be lost unnecessarily and

new synchronisations would be triggered immedi-

ately, thus increasing the overhead.

• The Procedures adopted Once a Region Becomes

Empty: The VNLayer does not preserve informa-

tion about the states of the VNs corresponding to

regions that become empty after the withdrawal of

all the PNs located in them. This degrades the per-

formance of the virtualization layer, thus causing

unnecessary delays in the recovery of the VNs.

Besides the above limitations (which have been

identiﬁed in MANET scenarios), the VNLayer re-

quires additional reﬁnements to improve its perfor-

mance in more restrictive and demanding mobile

communication environments, such as the pedes-

trian/vehicular/ mixed scenarios we want to explore

in our doctoral proposal.

3.2 Resource Sharing among Mobile

Devices

The idea of taking advantage of the resources avail-

able in the handheld devices that are located around

an on-move user is not new at all. Speciﬁcally, the ca-

pabilities that have gained more momentum are those

related to the networking resources. In this regard, the

so-called spontaneous networking has arisen in the

last years, where wireless mobile nodes opportunis-

tically exploit multi-hop ad-hoc paths toward peers to

share content and available resourcesin an impromptu

way. The idea is to take beneﬁt from the available

bandwidth in many handheld devices (which is of-

ten underutilized) to be shared with other peers in

current vicinity, thus better exploiting the increasing

availability of computing/memory/bandwidth-related

capabilities at portable wireless terminals. In this line,

we found the approach proposed in (Bellavista and

Giannelli, 2010) where a middleware named RAMP

(Real Ad-hoc Multi-hop Peer-to-peer) is described.

In particular, RAMP combines network-layer solu-

tions and application-layer approaches to support In-

ternet connectivity sharing in spontaneous networks.

Speciﬁcally, RAMP creates multi-hop paths toward

border nodes (i.e., nodes directly connected to the tra-

ditional Internet and offering part of their underuti-

lized connectivity to nearby peers), so that each node

can use the path currently deemed as the most suit-

able, e.g., because it provides largest bandwidth or

requires lowest power comsumption. This approach

leads to a signiﬁcant routing overhead when exploit-

ing different multi-hop heterogenous paths traversing

the same node.

At the application layer we found other ap-

proaches that go beyond the solutions proposed in

RAMP, which have been designed for vehicular ad-

hoc networks taking inspiration from BitTorrent-style

P2P ﬁle sharing systems (Nandan et al., 2005; Lee

et al., 2007; Chen and Chan, 2009; Lee et al., 2006;

Eriksson et al., 2008). The goal is not only to ex-

ploit the connectivity of one terminal from the rest

of devices, but to collaboratively download different

chunks of the same content during periods of con-

nectedness. All these application-layer protocols are

aimed at enabling (collaborative) downloads of con-

tents that typically are appealing to all (or most of) the

vehicles connected to the VANET. The contributions

that we are pursuing in SCC are located in a lower

layer, where the goal is to aggregate the connections

of several nodes in a transparent way, without con-

ditioning besides the usage of the downloaded con-

tents by those nodes (covering, e.g., scenarios where

the accessed information is useful for just one node

in the SAN). To this aim, we must envisage tranport-

layer solutions that deal with multiple connections

and multi-hop communications in diverse mobile ad-

hoc networks, by working on the top of our enhanced

virtualization layer, which, to the best of our knowl-

edge, is approach completely novel in literature.

Beyond the networking capabilities, in the vehicu-

lar environment it is also possible to ﬁnd new service

models that allow vehicles to share to each other on-

board storage facilities (STaaS: Storage as a Service),

computing power (CaaS: Computing as a Service),

services (about trafﬁc information, driver safety or

weather and road conditions) that are assembled from

the information collected by other vehicles (COaaS:

Collaboration as a Service), and advanced function-

alities related to provision of entertainment as a ser-

vice on the road (ENaaS: ENtertainment as a Ser-

vice) or taking of photos and recording of videos in

particular places and at speciﬁc times (PicWaaS: Pic-

tures on a Wheel as a Service), as described in (Arif

SporadicCloudComputingoveraVirtualizationLayer-ANewParadigmtoSupportMobileMulti-hopAd-hocNetworks

et al., 2012). These services can be deployed over

diverse vehicular clouds, ranging from static clouds

(which aggregate the capabilities of parked cars) and

semi-static clouds (involving vehicles stopped for a

moment because of a trafﬁc jam) to mobile clouds

(the most common option where a large amount of

vehicles travel on the road). The most sophisticated

approaches have been designed in static and semi-

static clouds, while the challenges derived from the

frequently changing topologies of mobile clouds have

not received the same attention (Gerla et al., 2014).

The goal of our proposal is to handle the mobility of

the nodes connected to our SANs (both pedestrians

and vehicles), by exploiting the virtualization reﬁne-

ments and the mechanisms envisaged in the SCC to

face the communication errors and data loss notice-

able in highly dynamic ad-hoc communication envi-

ronments (Arif et al., 2012).

4 METHODOLOGY AND STAGE

OF THE RESEARCH

After an in-depth review of the state-of-the-art, the

ﬁrst step to tackle our research problem has been the

development of a simulator (whose high-level design

is sketched in Section 4.1), which is aimed at vali-

dating the procedures of the VNLayer+ (Section 4.2)

and the “X”aaS service models of the SCC paradigm

(Section 4.3). While the simulator has been totally

implemented, our ongoing work is focused on the

foundations of the VNLayer+ and the speciﬁc mech-

anisms of some “X”aaS models of the SCC.

4.1 A SAN Simulator

Covering vehicular, pedestrian and mixed environ-

ments requires that our simulator (i) models the mo-

bility requiremens of each application scenario, and

(ii) deals with the communications among the mobile

nodes. To this aim, we have revised diverse pedes-

trian and vehicular mobility models deﬁned in lit-

erature (Sharma and Singh, 2013), with the goal of

selecting the ones that represent realistically the be-

haviours of the moving nodes that are connected to

our SANs. Regarding the pedestrians, as depicted

in Figure 2, we have chosen three different models

to generate diverse types of mobility traces, referred

both to individuals and groups.

• The Random Walk Mobility Model allows the

nodes to move randomly without restrictions (e.g.

a pedestrian walking a street in a city), so that the

destination, speed and direction are all chosen in-

dependently of other nodes.

• The Nomadic Community Mobility Model consid-

ers groups of nodes that collectively move from

one point to another. This model is especially ap-

propriate for scenarios where a group of pedes-

trians move together, but each individual could

also roam around a particular location individu-

ally (e.g. a group of tourists who visit together the

historical centre of a city). By adjusting the corre-

sponding parameters, it is possible to control how

far each node can roam from each reference point,

thus resulting into very realistic movements.

• Finally, we adopt the Reference Point Group Mo-

bility Model where each group is composed of a

number of members and one leader, so that the

movements of the leader determine the mobility

behaviour of the entire set (e.g in a mobility sce-

nario where a group of students visit a museum,

being guided by an expert).

As depicted in Figure 2, the above mobility mod-

els have been integrated via the existing simulation

tool MobiSim

, whose modular architecture allows to

easily add extra models and trace formats. Regarding

the generation of vehicular mobility traces, we have

resorted to SUMO

, due to the possibility of adding

new mobility models and submitting realistic (vehic-

ular) mobility traces in NS-2 format.

Also, we have decided to adopt NS-3 because

this simulator greatly improves NS-2 in terms of efﬁ-

ciency, memory management and kernel architecture,

besides making easier the integration of third-group

software and the deﬁnition of new mobility models

by using C++. Certainly, the models implemented in

NS-3 are too simple in order to fulﬁll a wide diver-

sity of mobility requirements. However, our simula-

tor overcomes this limitation thanks to the (pedestrian

and vehicular) mobility behaviours modeled by the

external simulators MobiSim and SUMO. As these

behaviours are modeled as NS-2 traces, NS-3 uses a

NS2MobilityHelper module to convert them to NS-

3 mobility events. As seen in Figure 2, NS-3 sup-

ports protocols such as UDP, TCP, IP and multiple

routing protocols for mobile ad-hoc networks. Con-

sidering our virtualization mechanisms requires to in-

clude two additional modules aimed at implementing

the virtual layer level (VNLayer+) and a virtualized

routing protocol grounded on it (VNRouting), thus

greatly improving the performance of the ad-hoc net-

work. Lastly, the lowest level hosts diverse versions

of the IEEE 802.11 protocol (such as IEEE 802.11p

speciﬁcally developed for vehicular networks).

http://www.masoudmoshref.com/old/myworks/

documentpages

http://sumo.sourceforge.net/

CLOSER2015-DoctoralConsortium

Figure 2: High-level design of our SAN simulator.

In conclusion, the exploitation of synergies among

MobiSim, SUMO and NS-3 makes it possible to eas-

ily deﬁne multiple and diverse mobility scenarios

where we will explore the potential of the SANs con-

tributed in our proposal.

4.2 The VNLayer+

As depicted in Figure 3, the VNLayer+ divides the ge-

ographical area of the SAN into regions (denoted as

cubes), so that a virtual node is located in each region.

Analogously to what we commented for the VNLayer

by Dolev et al., each virtual node is supported by sev-

eral moving nodes (both pedestrians and vehicles in

our scenarios), so that the virtual node stops to work

after all the physical nodes leave the region. Also,

there exist in each region leader and backup nodes in

order to maintain replicas of the virtualization-related

state information and the routing tables tackled by the

routing protocols working on the VNLayer+.

We have focused on the envisage of procedures

aimed to face the inefﬁciency sources identiﬁed in the

traditional VNLayer (recall Section 3.1). Next, we

sketch the ideas considered in our reﬁnements, whose

details can be found in (Bravo-Torres et al., 2015):

• First, we have developed a new leader election

procedure to prevent from the slow reaction of the

VNLayer to leader withdrawls, which impinged

heavily on the communications in scenarios of

high mobility, since the VNs were down during a

non-negligible portion of the average time that the

vehicles would remain in the respective region.

Brieﬂy, the leader election procedure is driven by

different types of events (message receipts, time-

outs and region changes), which take each PN

to multiple states. Our approach consists of re-

moving some of these states and reorganizing the

transitions between the remaining ones in order to

speed up the discovery of the new leader.

Besides, we are also interested in sophisticating

the election of VN leaders so that the role is dy-

namically transferred to the physical node that

is most likely to remain longest within the cor-

responding region (as inferred from information

coming from either the link layer or the applica-

tions layer).

• The backup designation procedure proposed in

the VNLayer needs improvements too. In particu-

lar, our goal is to ensure that the number of backup

nodes in a region stays, whenever possible, within

a given minimum (to guarantee the resilience of

the virtual nodes) and a given maximum (to avoid

excessive synchronisation overhead). To this aim,

our idea is that the leader reports the number of

backups in the region, so that other non-leaders

can learn whether they should offer themselves

to further support the VN. This way, becoming a

backup is no longer a fortuitous and autonomous

decision as in Dolev et al.’s approach, but rather

an informed and supportive one.

• Last, we have also developed procedures to im-

prove the management of empty regions designed

in the VNLayer. In this regard, our approach is

based on deﬁning a new table (named B

table)

whose entries contain the physical addresses of

the backup nodes along with its state (synchro-

nised or not). Speciﬁcally, B

table replicas are

stored in all the nodes of a region, which is the

key to avoid losing state information from the

upper layers when a newcomer assumes leader-

ship shortly after the previous leader has left. In

this scenario, the upstart (the newcomer) does not

have the state information of the virtual VN of the

region, but the synchronised backup do. Com-

bining the B table and the information from the

backup node, the upstart can start operating just

as well as the former leader in a very short time.

These mechanisms are complemented with other

procedures aimed to keep the structure of our

SANs stable, thus enabling that the communica-

SporadicCloudComputingoveraVirtualizationLayer-ANewParadigmtoSupportMobileMulti-hopAd-hocNetworks

Figure 3: Sporadic Ad-hoc Network (SAN) deployed in a generic mobility scenario involving pedestrians and vehicles.

tions and services over the ad-hoc networks are

not interrupted as users move. For that purpose,

when a virtual node is about to become inactive

(because there are very few devices in that re-

gion), the leader sends its buffered information to

the leader node of a neighbor region. This infor-

mation is stored until the original virtual node is

operative again (i.e. until new terminals enter that

region). At this moment, the just-restored virtual

node requests the information and processes it by

resorting to the capabilities provided by the new

terminals that are located in its region, thus pre-

venting from losing information as nodes move.

4.3 The SCC Paradigm

The devices connected to the SANS require coordina-

tion mechanisms to orchestrate the sharing and allo-

cation of their available resources (and even of extra

resources available from the Internet). To this aim, the

virtual nodes of the VNLayer+ must establish com-

munications to each other in order for the users’ de-

vices (i) to request resources to other terminals and

(ii) to advertise those that they are willing to pro-

vide to the SAN. We have designed a transport-layer

approach aimed at sharing and allocating resources,

on which the multiple “X”aaS services of our SCC

paradigm will be grounded. Each of these service

models will require particular reﬁnements (on which

we are focusing our ongoing research work) that will

be develop on the common substratrum described in

this section.

For our descriptions, we assume that the geo-

graphic area where the SAN has been deployed is

divided into regions (recall Figure 3), whose leaders

and backup nodes have been selected by the mecha-

nisms of the VNLayer+ mentioned in Section 4.2. In

this scenario, we suppose that a terminal connected to

the SAN (hereafter, application node) needs extra re-

sources for running an application and asks for them

to the sporadic cloud. This process is organized as

follows, as depicted in Figure 4:

• Firstly, the application node broadcasts a Re-

source Discovery Message (

MRDiscovery

) in its

region.

CLOSER2015-DoctoralConsortium

Figure 4: Messages exchanged among the application node

and the leaders of the regions identiﬁed within a SAN.

• Upon the reception of the

MRDiscovery

message,

the leader node of the region sends it to the leaders

of its adjacent regions. This process is repeated by

the remaining leaders until reaching the last re-

gion.

• In each region, the leader includes in the

MRDiscovery

message information about the re-

sources available in the devices supporting that

virtual node. The leader of the last region aggre-

gates all the responses and sends this information

back to the application node via a Resource Re-

ply Message (

MRReply

) through the leaders of the

intermediate regions.

• After receiving the

MRReply

message, the appli-

cation node distributes tasks among the virtual

nodes, considering the availability of resources

reported by the leaders of their respective re-

gions. The corresponding task assignment is noti-

ﬁed to each leader via a Resource Acknowledge-

ment Message (

MRAck

). This allocation process

changes as per the speciﬁc “X”aaS service de-

ployed, which at the same time depends on the

kind of resources requiredfor the application node

(computing, storing, networking...).

• Since virtual nodes might need to cooperate when

it comes to getting the information required by the

application node, each leader records the tasks to

be done by the rest of leaders. This information is

also reported to the backup nodes of each leader in

order to ensure a correct synchronization among

them, so that a backup can take the place of the

current leader when this node leaves the region.

• After receiving the

MRAck

message, each leader

distributes its task assigment among the physical

nodes of the region, by considering the capabili-

ties that these devices put at disposal of the SAN

at this moment. Once the terminals have ﬁnished

their job (which depends obviously on the capa-

bilities shared in each “X”aaS service), the leader

of this region is notiﬁed. This node ﬁnally informs

to the application node by sending a Completed

Task Message (

MCTask

• This way, the cooperation among the virtual nodes

enables to get the information required by the ap-

plication node. Depending on the application to

be runned in this node, our approach allows to

replicate this information in special virtual nodes

of the SAN, so that other application nodes can

also access it. Speciﬁcally, to this aim, we resort

to very stable virtual nodes which are supported

by a huge amount of connected devices that are

located, for example, in crowded squares or av-

enues and intersections with a lot of vehicular traf-

ﬁc. In order to access these contents, each node

just needs to send a Content Request Message

(

MCRequest

) to the leader of its region to trigger

the process. The information is ﬁnally received

via a Content Response Message (

MCResponse

Note that above descriptions are also valid for a

scenario where multiple application nodes ask for re-

sources simultaneously. In this case, the capabilities

available in each virtual node will be considered when

deciding about new requests, and, once the resources

provided by the terminals are about to ﬁnish, the re-

quests will be queued until the resources blocked by

other application nodes are released.

As introduced before, having developed the com-

mon substratrum of SCC, we need to envisage the

speciﬁc mechanisms required in each “X”aaS service.

Currently, we are working on the N(etworking)aaS

model with the goal of allowing the mobile nodes to

collaboratively download contents from the Internet

and to share them with the rest of members over the

SAN. To this aim, we require transport-layersolutions

that manage multiple connections over the multi-hop

ad-hoc network in a transparent way. These mech-

anisms should lead to signiﬁcant improvements in

terms of download time, thanks to the simultaneous

exploitation of the Internet connections providedfrom

the terminals of several individuals (e.g. if we have 3

terminals with a bandwidth of 1 Mbps each one to

download a content of 3 MB, the download would

last 3 seconds using just one connection, and 1 second

splitting the content in three chunks of 1 MB each one

and aggregating the 3 connections).

SporadicCloudComputingoveraVirtualizationLayer-ANewParadigmtoSupportMobileMulti-hopAd-hocNetworks

5 EXPECTED OUTCOME

In our doctoral proposal, wireless mobile nodes op-

portunistically harness multi-hop ad-hoc paths to ex-

ploit the availability of a great amount of (often un-

derutilized) resources that are provided by the termi-

nals of nearby “always-on” users in the context of the

SANs.

On the one hand, such SANs promote shared ex-

periences among potentially like-minded users who

happen to be close to each other, by relying on direct

or hop-by-hop ad-hoc communications. On the other

one, in the realm of these ad-hoc networks, our new

SCC paradigm enables the deployment of multiple

context-aware applications and “X”aaS service mod-

els whose novelty is grounded on two features. First,

the fact of working with static virtual nodes instead

of mobile real nodes, which makes easier the routing

tasks and ensures reliable and stable communication

environments. Second, the possibility of bringing to-

gether in a transparent way (i) the resources provided

by each terminal that is connected to the SAN and

(ii) the capabilities of the traditional MCC (if avail-

able) in diverse ad-hoc mobility scenarios, working at

the transport layer and on the top of the virtualization

mechanisms.

To put it from another angle, the resource shar-

ing and allocation pursued in our SCC allows to im-

prove the experience of each individual thanks to the

capabilities contributed by the rest of members of the

SAN, notwithstanding their mobility. This way, the

users might, for instance, enjoy connectivity of the

Internet and even reduce download times by aggre-

gating the 3G/4G connections of other terminals, use

extra storage space (in these devices or even in the

cloud) and also exploit additional computational re-

sources provided by more powerful terminals in or-

der to run, for example, (demanding) personalization

strategies. Such strategies would allow to provide the

SAN users with contents of their interest which might

havebeen proactivelyselected as per their preferences

(and even collected, enhanced and shared by other in-

dividuals) within tailor-made sporadic communities.

ACKNOWLEDGEMENT

This work is being funded by the Ministerio de Edu-

caci´on, Cultura y Deporte (Gobierno de Espa˜na) re-

search project TIN2013-42774-R.

REFERENCES

Arif, S., Olariu, S., Wang, J., Yan, G., and Khalil, I.

(2012). Datacenters at the airport: reasoning about

time-dependent parking lot occupancy. IEEE Trans-

actions on Parallel and Distributed Systems, 23:2067–

2080.

Bellavista, P. and Giannelli, C. (2010). Internet connectiv-

ity sharing in multi-path spontaneous networks: Com-

paring and integrating network- and application-layer

approaches. In MOBILWARE’10, 3rd International

ICST Conference on MOBILe Wireless MiddleWARE,

Operating Systems, and Applications, pages 84–99.

Bravo-Torres, J. F., L´opez-Nores, M., Blanco-Fern´andez,

Y., Pazos-Arias, J. J., and Ord´o˜nez-Morales, E. F.

(2015). Vanetlayer: A virtualization layer support-

ing access to web contents from within vehicular net-

works. Journal of Computational Science, In press.

Chen, B. B. and Chan, M. C. (2009). MobTorrent: A frame-

work for mobile internet access from vehicles. In

INFOCOM’09, 24TH IEEE Conference on Computer

Communications, pages 1404–1412.

Dinh, H. T., Lee, C., Niyato, D., and Wang, P. (2013). A

survey of mobile cloud computing: architecture, ap-

plications, and approaches. Wireless Communications

and Mobile Computing, 13(18):1587–1611.

Dolev, S., Gilbert, S., Lynch, N. A., E., E. S., Shvartsman,

A. A., and Welch, J. L. (2004). Virtual mobile nodes

for mobile ad hoc networks. Distributed Computing,

3274:230–244.

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

Cabernet: Vehicular content delivery using WiFi. In

MobiCom’08, 14th ACM International Conference on

Mobile Computing and Networking, pages 199–210.

Gerla, M. (2012). Vehicular cloud computing. In MED-

HOC-NET’12, 11th Annual Mediterranean Ad-Hoc

Networking Workshop. IEEE Press.

Gerla, M., Wu, C., Pau, G., and Zhu, X. (2014). Content

distribution in VANETs. Vehicular Communications,

1(1):23–12.

Kuss, D. J. and Grifﬁths, M. D. (2011). Excessive online

social networking: Can adolescents become addicted

to facebook. Education and Heath, 29(4):68–71.

Lee, K., Lee, S. H., Cheung, R., Lee, U., and Gerla, M.

(2007). First experience with CarTorrent in a real ve-

hicular ad-hoc network testbed. In 2007 Mobile Net-

working for Vehicular Environments, pages 109–114.

Lee, U., Park, J. S., Yeh, J., Pau, G., and Gerla, M. (2006).

Code torrent: Content distribution using network cod-

ing in VANET. In DIWANS’06, Workshop on Depend-

ability Issues in Wireless Ad Hoc Networks and Sensor

Networks 2006 (part of Mobicom 2006).

Nandan, A., Das, S., Pau, G., Gerla, M., and Sanadidi, M.

(2005). Cooperative downloading in vehicular ad-hoc

networks. In WONS’05, 2nd International Confer-

ence on Wireless On-Demand Network Systems and

Services, pages 32–41.

Olariu, S., Hristov, T., and Yan, G. (2013). The next

paradigm shift: from vehicular networks to vehicular

CLOSER2015-DoctoralConsortium

clouds. In Basagni, S., Giordano, S., and Stojmen-

ovic, I., editors, Mobile ad-hoc networking: cutting

edge directions. John Wiley & Sons.

Przybylski, A. K., Murayama, K., DeHaan, C. R., and Glad-

well, V. (2013). Motivational, emotional, and behav-

ioral correlates of fear of missing out. Computers in

Human Behavior, 29(4):1841–1848.

Sharma, A. and Singh, E. J. (2013). Mobility models for

manet: Mathematical perspective. International Jour-

nal of Advanced Research in Engineering and Applied

Sciences, 2(5):59–68.

SporadicCloudComputingoveraVirtualizationLayer-ANewParadigmtoSupportMobileMulti-hopAd-hocNetworks