CoP3D: CONTEXT-AWARE OVERLAY TREE FOR

CONTENT-BASED CONTROL SYSTEMS

∗

Mauro Caporuscio

Dipartimento di Elettronica de Informazione, Politecnico di Milano, Italy

Alfredo Navarra

Dipartimento di Matematica e Informatica, Universit`a degli Studi di Perugia, Italy

Keywords:

Content-based publish/Subscribe system, Sensor networks, Distributed computing, Overlay network.

Abstract:

Publish/Subscribe systems are nowadays largely accepted for coordinating applications over wide-area net-

works, whereas they still suffer when applied to mobile environments. In fact, the typical tree overlay network

implementation of the event service, which provides scalability in the context of wide-area networks, fails

when a run-time reconﬁguration is needed. This might happen when dealing with the dynamic and mobility

characteristics inherent to the pervasive environments. In this proposal, we address such an issue by envi-

sioning a context-aware approach in order to build and maintain the dispatching tree overlay with respect to

the context sensed through the environment. The resulting distributed control system optimizes on the energy

consumption aspect, which is the key cost measure for a mobile environment such as sensor networks.

1 INTRODUCTION

Continuous improvement of wireless network tech-

nologies and the miniaturization of electronic equip-

ments are de facto enabling the exploitation of wire-

less sensor networks as distributed control systems

for monitoring data within physical environments. In

particular, a sensor network consists of a number of

sensor nodes randomly disseminated within the envi-

ronment. This requires for protocols and algorithms

that let nodes to self-organize and cooperate with each

other.

In this context, content-based publish/subscribe

system is considered. Indeed, it is a well suited com-

munication infrastructure for dealing with distributed

applications over wireless networks. In fact, the loose

coupling characteristic of publish/subscribe together

with content-based routing protocol, where the mes-

sage routing is determined by the interests of the re-

ceiver rather than by the explicit destination address,

allow for dealing with the dynamism of such environ-

ments (Carzaniga and Wolf, 2001).

As depicted in Figure 1, a content-based pub-

lish/subscribe system is composed of two main en-

tities: clients and servers. Servers are interconnected

∗

This research has been partially funded by the Eu-

ropean Commission, Programme IDEAS-ERC, Project

227077-SMScom.

Figure 1: Tree overlay implementation of the event service.

in a distributed network, referred to as event service,

and provide clients with access points offering an ex-

tended publish/subscribe interface. Clients are of two

kinds: (i) publishers use the access points to publish

events and (ii) subscribers use the access points to

subscribe for events of interest by supplying a pred-

icate, called subscription. A subscription is applied

to the content of events and it allows subscribers to

select the events they are interested in. The event ser-

vice is responsible for selecting and delivering events

of interest to subscribers via the access points. The

idea of a publish/subscribe system is quite mature,

and a fair number and variety of publish/subscribe

systems have been proposed, including research pro-

totypes (Carzaniga et al., 2001; Cugola et al., 2001;

305

Caporuscio M. and Navarra A. (2009).

CoP3D: CONTEXT-AWARE OVERLAY TREE FOR CONTENT-BASED CONTROL SYSTEMS.

In Proceedings of the 6th International Conference on Informatics in Control, Automation and Robotics - Intelligent Control Systems and Optimization,

pages 305-310

DOI: 10.5220/0002247703050310

 SciTePress

Pietzuch and Bacon., 2002; Meier and Cahill, 2002),

commercial products (TIBCO Inc., Palo Alto, CA,

1996), and attempts of standardization (Object Man-

agement Group (OMG), 2004; Sun Microsystems,

Inc., Mountain View, California, 1999).

The typical implementation of the event service

is a tree overlay network (as shown in Figure 1),

which provides scalability in the context of wide-

area networks (e.g., Siena (Carzaniga et al., 2001)

and Elvin (Segall and Arnold., 1997)). In fact, hav-

ing a unique path between any two nodes simpliﬁes

both the matching algorithm and the routing scheme,

and avoids network ﬂooding. Even though such an

overlay topology has been successfully exploited in

ad-hoc environments (see e.g., multicast tree (Royer

and Perkins, 2000)) there are issues concerning its

application to publish/subscribe systems in the con-

text of mobile environments. In fact, the high degree

of dynamics inherent to sensor networks requires to

continuously reconﬁgure the event service and, con-

sequently, to reconﬁgure data structures and routing

tables. Such structures, distributed through the event

service, are created by considering the interest ex-

pressed by the clients and evolve at runtime when sub-

scriptions are added/removed/modiﬁed. This makes

the reconﬁguration process expensive (in terms of

both message exchanges and computational opera-

tions) and hard to accomplish.

Many works have been published so far con-

cerning publish/subscribe systems in mobile environ-

ments, such as (Huang and Garcia-Molina, 2004;

Fiege et al., 2003). However, most of them focus on

the clients mobility while the event service remains

stable (Caporuscio et al., 2003). On the other hand,

in (Huang and Garcia-Molina, 2003) and (Mottola

et al., 2008), the authors concentrate on the event ser-

vice topology reconﬁguration in the ﬁeld of ad-hoc

networks.

This paper deals with publish/subscribe systems

over sensor network by proposing a context-aware

technique to build and maintain the dispatching tree

overlay with respect to the context sensed through the

environment. Our proposal is to exploit the Connec-

tionless Probabilistic (CoP) protocol (McCann et al.,

2005), originally designed for sensor networks, in the

context of publish/subscribe systems. The CoP proto-

col assumes battery powered mobile devices that in-

teract with each other in order to establish a virtual

infrastructure for routing purposes. Namely, a virtual

grid infrastructure is built by considering the physical

position of the devices. Further, the protocol takes

care of the energy consumption by beneﬁting from

mobility and network dynamics.

We envision an extension of the CoP protocol in

order to build a rooted tree on top of the virtual grid.

This would realize a new protocol, called CoP3D,

where the tree is used to accomplish publish/subscribe

transmissions (i.e., the overlay network), whereas

the underlying grid maintains its original aim. The

main property of the proposed CoP3D protocol is

that it builds the overlay network taking into account

context information sensed through the environment.

Namely, the devices representing the tree nodes will

be selected according to their physical location, avail-

able energy, computational resources and expected

mobility rate.

The paper is organized as follows. Section 2 de-

scribes the CoP protocol and the virtual grid construc-

tion. Section 3 presents CoP3D and describes how the

tree overlay is built with respect to the sensed context.

Section 4 points out interesting peculiarities of the

considered context that must be carefully addressed.

Section 5 discusses future work.

2 CoP: THE VIRTUAL GRID

INFRASTRUCTURE

In this section, we outline the CoP protocol (McCann

et al., 2005), which aims to manage communication

for sensor networks. We have chosen this protocol

since it efﬁciently performs in mobile ad-hoc environ-

ments and well ﬁts our purposes.

We assume the devices can move over some given

area A. From now on we refer to devices also by

nodes. As it will be clear soon, our approach can be

adapted to any kind of desired shape for A. It may be

suitable for any building like ofﬁces, airports, com-

panies and so forth where the network has to be de-

ployed. For the sake of clarity, A is assumed to be sim-

ply a square area of side d. At the border of A there

can be some ﬁxed infrastructure, if required, that sup-

ports the survivability of the network. In the context

of publish/subscribe systems, whenever a publisher

wants to provide a service on the network, it has to

inform the network about its publication. In the wired

context this message was routed along a ﬁxed tree re-

sponsible of the matching among publishers and sub-

scribers. Since in our wireless and mobile context we

cannot have such an infrastructure, the basic idea is to

build a virtual one along which messages are routed.

Virtual infrastructure means that activenodes have

to take care somehow of its feasibility and mainte-

nance. The only thing that each node needs to know

in order to participate in the network, by means of

the CoP protocol, is its own physical position with re-

spect to the area A. This can be achieved by means of

either devices powered by GPS systems or some in-

ICINCO 2009 - 6th International Conference on Informatics in Control, Automation and Robotics

306

sink #1

sink #2

Storage Device

Centralized

d/u

Figure 2: The area of interest covered by a virtual grid in-

frastructure of grid unit

. Nodes inside each circle of ra-

dius ds compete to represent the corresponding grid node.

stalled beacon external nodes that inform about rela-

tive positions (the mentioned infrastructure at the bor-

der of A), or other positioning techniques (see for in-

stance, (Capkun et al., 2006; Caruso et al., 2005)).

According to the expected population inside A, CoP

builds a virtual infrastructure. Namely, a square grid

G covering A is considered with grid unit d/u (see

Figure 2), where u is determined according to the den-

sity of the nodes. From the “balls into bins” theory

(see (Raab and Steger, 1998; McCann et al., 2005)),

by choosing an appropriate distortion parameter ds,

according to the density of the nodes in the area of in-

terest, we can compute our desired value u in such a

way that for each grid node there will be some device

inside the circle of radius ds centered on it with high

probability. We can choose for instance u =

3ds

. Note

that u can change according to the established density

of the network over the time. Its current value can be

part of the beacon message sent by the external border

nodes.

Whenever a device in A starts its interaction with

the network or just moves from its current position,

it is aware about its location with respect to A and

hence to G. This allows for determining whether it

can be elected as representative of a grid node. For

this step of the CoP virtual grid construction, we refer

to standard leader election strategies suitably adopted

in wireless environments like ad-hoc and sensors net-

works (see (Malpani et al., 2000)).

In doing so, at the expenses of some communi-

cation needed for the evaluation of a suitable value

for u and the local leader elections, CoP builds a vir-

tual grid G that can be used in order to correctly route

desired communications. G covers all the given area

A. One of the most important properties of such a

construction is that it does not depend on the current

nodes acting as leaders but only on the density of the

nodes. The CoP protocol was originally applied in the

u = 0 means there are not enough devices in the area to

build the network.

ﬁeld of sensor networks where sensed data needed to

be route outside A to a ﬁxed sink. Hence, the routing

was performed in a multi-hop fashion over the grid.

The main steps that must be performed by a sensor in

order to participate to the CoP protocol can be sum-

marized as follows.

1. It discoversits position according to the used tech-

nology, the area covered by the network and the

size of the grid needed for the routing

2. If the received value for u is zero then it just waits

3. According to the previous information it evaluates

its position with respect to the grid and decides

whether it can represent any grid node or not. If

not, it just assumes there is someone representing

the closest grid node to its position and whenever

needs to send a message it delegates such a node

for the correct routing. If yes, it checks whether

there is someone else playing or not that role

4. If it represents a grid node then it has to take care

of the associated trafﬁc that has to be route

5. One hop of the communication is made over the

grid in such a way that the transmission can be

received inside the whole circular area associated

to the target grid nodes. This is due to the fact

that a transmitting node does not know the exact

position of its neighbors but it only assumes that

someone is inside the circle of radius ds surround-

ing the target grid node

6. Whenever a device changes its position, i.e., ei-

ther enters or exits a circular area, it has to change

its role accordingly. If it was responsible for the

routing, it has to take care to send the needed in-

formation to at least one of the current candidate

to become its successor

7. If a device is running out of energy, it has to be-

come a passive node even though still continues

to receive its desired communications until it can.

Starting from this routing protocol and infras-

tructure, next section shows how this can be modi-

ﬁed and suitably adapted in the case of mobile pub-

lish/subscriber systems.

3 CoP3D: THE TREE OVERLAY

NETWORK

In this section, we describe our proposal concerning

the management of a mobile publish/subscriber sys-

tem. In particular,we exploit the CoP virtual grid con-

struction in order to obtain a tree overlay implement-

ing the event service. The main differences with re-

CoP3D: CONTEXT-AWARE OVERLAY TREE FOR CONTENT-BASED CONTROL SYSTEMS

307

Figure 3: The tree construction based on the virtual grid

infrastructure.

spect to a typical structured environment can be sum-

marized as follows:

(a) Energy Efﬁciency: mobile devices have scarce

power capacity and they can be required to stay

alive for long periods without any support

(b) Scalability: it is desirable that any solution to

manage such type of network would easily scale

according to the number of devices and the corre-

sponding area. Due to the mobility feature, man-

ual deployment of devices is just not feasible

also usually designed to cope with fault-tolerance

issues, in a mobile environment faults are much

more frequent. A device can easily disappear

from the network due to either its movement or

its low power level

(d) Absence of the Infrastructure: Due to their move-

ments or to some handoff strategy, a given device

is not always linked to the same set of neighbors

at different time. This implies that assumptions

about the topology of the network cannot be done

unless they are straightforward from the area of

interest (for instance networks along a street can-

not differ too much from a line). The only ﬁxed

infrastructure that can be assumed can reside at

the border of the area of interest

As shown in Figure 3, virtual grid nodes become

the leaves of the tree overlay. Over the basic vir-

tual grid infrastructure previously described, we build

several levels of grids. What in the typical pub-

lish/subscribe system was the routing tree now is a

4-ary tree rootedat the center of the grid. Dividing the

grid into four sub-grids we iteratively discover other

4 centers (one for each subgrid) to which the previous

one is virtually connected. We iterate such a process

2log

times in order to obtain a full coverage of the

grid by means of a 4-ary tree of logarithmic height

with respect to the number of grid nodes (see Fig-

ure 3). Starting from the bottom layer, once the basic

virtual grid is built, all the nodes also know their loca-

tion with respect to the overlay tree and hence, each

one can play its corresponding role. By referring to

Figure 3, for instance, node v knows that in the over-

lay tree it plays the role of parent for node z.

As it was for the virtual grid construction, many

devices may cleverly compete in a leader election

in order to become nodes of the tree overlay, hence

responsible for routing publish/subscribe events and

subscriptions. In particular, such an election is per-

formed by considering context information. These

include: (i) available energy and physical location to

be compliant with the CoP protocol (see Section 2),

(ii) mobility rate since the more a node is stable the

more a node is qualiﬁed to play the router role, and

(iii) computational resources since powerful nodes

can better accomplish routing tasks.

In this way, the devices representing the virtual

grid nodes of the 4-ary tree are responsible for the

routing of the messages to realize the desired pub-

lish/subscribe system. Such nodes will be clearly

more loaded than others and hence they will spend

more energy. In order to distribute the energy con-

sumption, changes to the grid structure can be done

by means of shifting procedures or changing the grid

unit u. It can also be implicitly obtained by means of

a certain mobility ratio. Whenever a device changes

its position, in fact, it has to evaluate if its role is

also changing according to the distortion parameter

ds. When a node moves out from the virtual circu-

lar area of radius ds deﬁned around each grid node, it

cannot be anymore representative of such a grid node

hence it has to delegate someone else for doing its

job. This means that mobility can play a central role

in this routing process since it implies a more uni-

form distribution of the energy consumption. Once

the tree overlay network is built, the event service can

act as in the case of the typical structured environ-

ment. This allows us to apply such an approach to any

publish/subscribe system that relies on a tree overlay

network.

Concerning the properties outlined at the begin-

ning of this section, we want to point out how our

approach is suitable with respect to them. For (a) it

is worth to note that the only resource of the devices

is their energy. Moreover, both the virtual grid infras-

tructure and the tree overlay are heavily dependent on

the devices survivability. On the other hand, this is the

only resource that we can rely on in order to avoid the

installation of typical structured environments. The

tree overlay network construction is quite lightweight

since it requires only some basic calculations. Fur-

thermore the routing is well spread among all the de-

vices and it is made in a multi-hop fashion. It takes

ICINCO 2009 - 6th International Conference on Informatics in Control, Automation and Robotics

308

into account both the maximum distance covered by

means of a transmission (hence the maximum energy

spent) and the time needed to deliver a message (that

in the worst case requires only 4log

steps). As last

remark, our strategy makes use of the mobility rate as

a mean to save energy since roles change accordingly.

Concerning (b) and (c), our approach is easily

scalable. It is in its nature to adapt the network cov-

erage based on the density of the devices with respect

to the area that must be served. It also suitably copes

with fault-toleranceissues. In fact, devicesmight con-

tinuously appear (disappear) in (from) the network

without breaking virtual connections up to a certain

threshold. Once reached such a threshold, communi-

cations among the devices become unfeasible.

The last property (d) is easily solved by means of

the virtual grid infrastructure, hence avoiding the in-

stallation of both structured environments and more

powerful devices.

4 FURTHER KEY-POINTS

We now focus the attention on three main properties

that must be carefully addressed in our approach.

(i) As it was for structured event services, it is ev-

ident that starting from the root and descending until

the leaves, nodes are differently loaded in terms of

routing messages. As already outlined, one mean to

avoid this unbalanced energy consumption might de-

pend on the mobility rate or the shifting procedure.

On the other hand it is worth noting that our construc-

tion can be easily modiﬁed in the case we prefer to

not deploy the root at the center of the grid since we

may have information about the distribution of the de-

vices. The previous discussion is in fact based on the

assumption of a uniform distribution of the devices in-

side A. Consequently, it is easy to understand that the

area of interest can be subdivided into several areas

according to the expected density, hence applying our

methodology to construct different subtree overlays.

(ii) Another very important issue that must be ad-

dressed before practically apply our method concerns

the maximum distance that devices’ transmissions can

cover. With our construction, in fact, not only the

root is the most loaded node in terms of communica-

tions but also it is required to perform the most distant

transmissions overall the network. In order to avoid

such a situation and to better balance the energy con-

sumption among the nodes of the network, we may

think about a reversed structure. This implies that the

more a node is far from the root (with respect to the

tree overlay), the more must be its transmission dis-

tance. In Figure 4 it is shown how the tree overlay

Figure 4: The tree construction based on the reversed vir-

tual grid infrastructure. The aim is to better balance energy

consumption among the devices.

may become. In any case, when a device belongs to a

layer in which its power cannot cope with the required

transmission distance, it does not mean that it cannot

perform the required communications. In the header

of the messages can be added information in such a

way that a long transmission, in terms of covered dis-

tance, can be emulated by multi-hop transmissions of

shorter range. This recalls what was used in the ﬁeld

of ATM networks by the concepts of Virtual Channel

and Virtual Path communications (see e.g., (Flammini

and Navarra, 2009)). It is not in the aim of this paper

to go into details of such discussion but we want only

point out that the underlying virtual grid structure at

each layer can be used for this purpose.

(iii) Finally, CoP was based on another assump-

tion for which some communications can be wasted

due to the absence of a sensor representative of some

virtual grid node. According to the parameter ds, such

an event can happen rarely but still there is the pos-

sibility. Since in our context the tree overlay must

be somehow guaranteed until it is feasible, we need

some strategy to cope with this possibility. Due to the

underlying virtual grid structure, in fact, the transmis-

sions directed to a missing node can be easily rerouted

to some other grid path. For this issue we can re-

fer to (Mostarda and Navarra, 2008) where CoP was

modiﬁed in order to address security issues. Clearly,

at the expenses of a higher energy consumption, such

situations can be addressed.

5 FUTURE WORK

In this paper, we have envisioned a new approach to

achieve content-based publish/subscribe control sys-

tems in the context of sensor networks. In particular,

we have proposed a context-aware technique to build

and maintain the tree overlay network for communi-

cation purposes. Speciﬁcally, our approach builds on

CoP3D: CONTEXT-AWARE OVERLAY TREE FOR CONTENT-BASED CONTROL SYSTEMS

309

the CoP protocol and takes into account context in-

formation sensed through the environment. In fact,

the devices representing the tree nodes are chosen

by considering their physical location, available en-

ergy, computational resources and expected mobility

rate. Once the tree overlay network is built, the pub-

lish/subscribe system can act as in the case of the typi-

cal structured environment. This allows to apply such

an approach to any publish/subscribe system that re-

lies on a tree overlay network. It is of main interest

to study both the theoretical feasibility and the effec-

tiveness of our proposal. Moreover, our methodol-

ogy might be also applied to other distributed systems

rather than just the publish/subscribe ones (e.g., peer-

to-peer) and to build other topology overlay networks.

REFERENCES

Capkun, S., Cagalj, M., and Srivastava, M. (2006). Se-

cure localization with hidden and mobile base sta-

tions. In Proc. of the 25

Annual Joint Conf. of the

IEEE Computer and Communications Societies (IN-

FOCOM). IEEE Computer Society.

Caporuscio, M., Carzaniga, A., and Wolf, A. L. (2003). De-

sign and evaluation of a support service for mobile,

wireless publish/subscribe applications. IEEE Trans-

actions on Software Engineering, 29(12):1059–1071.

Caruso, A., Chessa, S., De, S., and Urpi, A. (2005). GPS

free coordinate assignment and routing in wireless

sensor networks. In Proc. of the 24

Annual Joint

Conf. of the IEEE Computer and Communications So-

cieties (INFOCOM), pages 150–160. IEEE Computer

Society.

Carzaniga, A., Rosenblum, D. S., and Wolf, A. L. (2001).

Design and evaluation of a wide-area event notiﬁca-

tion service. ACM Transactions on Computer Systems,

19(3):332–383.

Carzaniga, A. and Wolf, A. L. (2001). Content-based net-

working: A new communication infrastructure. In

NSF Workshop on an Infrastructure for Mobile and

Wireless Systems, number 2538 in Lecture Notes in

Computer Science, pages 59–68, Scottsdale, Arizona.

Springer-Verlag.

Cugola, G., Nitto, E. D., and Fuggetta, A. (2001). The

JEDI event-based infrastructure and its application to

the development of the OPSS WFMS. IEEE Transac-

tions on Software Engineering, 27(9):827–850.

Fiege, L., Grtner, F. C., Kasten, O., and Zeidler, A.

(2003). Supporting mobility in content-based pub-

lish/subscribe middleware. In Lecture Notes in Com-

puter Science, volume 2672, pages 103 – 122.

Flammini, M. and Navarra, A. (2009). Layouts for mobil-

ity management in wireless atm networks. Discrete

Applied Mathematics (DAM), 157(1):98–111.

Huang, Y. and Garcia-Molina, H. (2003). Publish/subscribe

tree construction in wireless ad-hoc networks. In Proc.

of the 4

Int. Conf. on Mobile Data Management

(MDM), pages 122–140.

Huang, Y. and Garcia-Molina, H. (2004). Publish/subscribe

in a mobile environment. Wireless Networks,

10(6):643–652.

Malpani, N., Welch, J. L., and Vaidya, N. H. (2000). Leader

election algorithms for mobile ad hoc networks. In

Proc. of ACM Joint Work. on Foundations of Mobile

Computing (DIALM-POMC).

McCann, J. A., Navarra, A., and Papadopoulos, A. A.

(2005). Connectionless Probabilistic (CoP) routing:

an efﬁcient protocol for Mobile Wireless Ad-Hoc

Sensor Networks. In Proc. of the 24

IEEE Int.

Performance Computing and Communications Conf.

(IPCCC), pages 73–77.

Meier, R. and Cahill, V. (2002). Steam: Event-based mid-

dleware for wireless ad hoc networks. In Proc. of

the 1

Int. Work. on Distributed Event-Based Systems

(DEBS), Vienna, Austria.

Mostarda, L. and Navarra, A. (2008). Distributed idss

for enhancing security in mobile wireless sensor net-

works. Int. Journal of Distributed Sensor Networks

(IJDSN), 4(2):83–109.

Mottola, L., Cugola, G., and Picco, G. P. (2008). A self-

reparing tree topology enabling content-based routing

in mobile ad hoc networks. Transaction on Mobile

Computing, 7(8):946–960.

Object Management Group (OMG) (2004). Notiﬁcation

service version 1.1.

Pietzuch, P. R. and Bacon., J. (2002). Hermes: A dis-

tributed event-based middleware architecture. In Proc.

of the 1

Int. Work. on Distributed Event-Based Sys-

tems (DEBS), pages 611–618, Vienna, Austria.

Raab, M. and Steger, A. (1998). “Balls into bins” - a sim-

ple and tight analysis. In Proc. of the 2

Int. Work.

on Randomization and Approximation Techniques in

Computer Science (RANDOM), pages 159–170.

Royer, E. and Perkins, C. (2000). Multicast Ad hoc On-

Demand Distance Vector (MAODV) Routing. IETF,

Internet Draft: draft-ietf-manet-maodv-00.txt.

Segall, B. and Arnold., D. (1997). Elvin has left the build-

ing: A publish/subscribe notiﬁcation service with

quenching. In Proc. of the Australian UNIX User

Group, pages 243–255, Brisbane, Australia.

Sun Microsystems, Inc., Mountain View, California (1999).

Java message service.

TIBCO Inc., Palo Alto, CA (1996). TIBR+: a WAN router

for global data distribution.

ICINCO 2009 - 6th International Conference on Informatics in Control, Automation and Robotics

310