REAL-TIME DATABASES FOR SENSOR NETWORKS

Pedro Fernandes R. Neto

Departamento de Inform

atica, Universidade do Estado do Rio Grande do Norte(UERN), Mossor

o, RN

Maria L. B. Perkusich

Departamento de Inform

atica e Estat

ıstica, Universidade Cat

olica de Pernambuco, Recife, PE

Angelo Perkusich

Departamento de Engenharia El

etrica, Universidade Federal de Campina Grande,

Campina Grande, PB, Caixa Postal 10105 - CEP 58109-970

Keywords:

Real-Time Databases, Petri Nets, Embedded Systems, Sensor Networks.

Abstract:

In the last years, the demand of embedded systems has been increased. Also, due to the increasing compe-

tition among different kind of companies, such as cellular phone, automobiles and industrial automation, the

requirements for such systems are getting more complex. However, the data storage and processing tech-

niques, for these environments, are insufﬁcient for the new requirements. In this paper, we develop a model

for the integration of real-time database technology with an embedded sensor network systems, to tackle such

deﬁciencies. Coloured Petri nets are the mathematical framework used for the modelling and analysis.

1 INTRODUCTION

In the last years, the demand of embedded systems

has been increased. Also, due to the increasing com-

petition among different kind of companies, such as

cellular phone, automobiles and industrial automa-

tion, the requirements for such systems are getting

more complex. Each one of them aggregate more

functionalities to embedded devices, as well as dif-

ferent processing capabilities for the products. How-

ever, one of the main problems to deal with due to

the increase in the functionalities and complexity of

these systems, results in an increase of the amount

of the data. Such data must be efﬁciently man-

aged and stored considering speciﬁc non-functional

requirements of the application, such as, for example,

size and power consumption.

Nevertheless, the techniques that exist for process-

ing and data storage in embedded devices are limited

and, therefore, they are insufﬁcient in face of the new

requirements of the applications. Aiming at assisting

such limitations, some commercial databases man-

agement systems (DBMS) already are being adapted

for embedded environments. However, as pointed

out in (Tesanovic et al., 2002), the a commercial

DBMS (i) does not execute normally in embedded op-

erational systems, (ii) does not support the database

properties (ACID: Atomicity, Consistency, Isolation

and Durability), (iii) do not minimize the system

resources use, and (iv) do not assure time require-

ments. So, it can be impracticable to use conventional

databases in embedded systems due to the particular

requirements of such systems.

Another observation about the correctness of these

systems is that while the correctness in the database

depends on the logical consistency of the data and

the transactions, for embedded systems the correct-

ness depends, also, on timing constraints. Due to the

need of timing constraints real-time database systems

(RTDB) can be applied in the context of embedded

systems, in order to satisfy them (Tesanovic et al.,

2002; Hansson, 2002).

In the context of this paper we consider sensor

networks as the embedded system case study. In

most cases, such sensor networks are build using

preprogrammed devices that send data for a central

database, where the data is stored for off-line analy-

sis and queries (Chen et al., 2001). This approach has

some limitations, such as communication overhead,

and power consumption in the case of wireless sensor

networks. In (Chen et al., 2001; Bonnet and Seshadri,

2000) it is considered a new architecture, where the

environment can be seen as a distributed database sys-

tem, formed by a server database and each component

is a local database.

However, still exist many subjects to be investi-

gated in this context. In this paper, we develop a

model for the integration of real-time database tech-

nology with an embedded sensor network systems.

Coloured Petri nets are the mathematical framework

used for the modelling and analysis (Jensen, 1997).

Also, we used a set of computational tools for editing

599

Fernandes R. Neto P., L. B. Perkusich M. and Perkusich A. (2004).

REAL-TIME DATABASES FOR SENSOR NETWORKS.

In Proceedings of the Sixth International Conference on Enterprise Information Systems, pages 599-603

DOI: 10.5220/0002644505990603

 SciTePress

and analyzing CPNs named Design/CPN (Jensen and

al, 1999). We have used the resulting model consid-

ering logical and timing constraints for both data and

transactions.

This paper is organized as follow: in Section 2 we

describe the sensor networks and its limitations due to

the existing approach. We show, also, characteristics

of the new architecture, to (Chen et al., 2001; Bonnet

and Seshadri, 2000), for these networks. In Section 3,

we detail the modelling for a real-time database appli-

cation for the sensor network. In Section 4 we show

the model analysis. Finally, in Section 5 conclusions

are presented.

2 SENSOR NETWORKS

The modern sensors are capable of storing and pro-

cessing local data, as well as transferring these data.

Thus, the processing can be made in the sensor net-

work, reducing the use of energy and the data traf-

ﬁc and, consequently, increasing the network life time

(Bonnet et al., 2000).

A sensor network consists on a great device num-

ber connected through communication interfaces that

can be communicated between itself through network

protocols. Each sensor have limited capacity of stor-

age and processing, or either, a sensor has a purpose-

general CPU to process and storage small space to

save programs code and data.

The sensors are not always ﬁxed in an infrastruc-

ture, being power through batteries, making the en-

ergy economy an important task. The communication

consumes much more energy than processing, becom-

ing attractive to reduce the amount of the data ﬂow

between knots by local processing. One another con-

sideration in sensor networks is that knots can inhabit

in hostile environments, what requires a robust sys-

tem for the fast recovery case happens imperfections

(Bonnet and Seshadri, 2000; Bonnet et al., 2000).

The sensors transactions possess time labels. Its

values must be frequently update, since sensors data

become invalid due to the timing constraints. Long-

running query, that periodically recomputation the

sensors transactions results, are a possibility to keep

these update results.

In sensor network, users typically ask three kinds

of queries, (Bonnet and Seshadri, 2000): Histori-

cal queries: these are typically aggregate queries

over historical data obtained from the device net-

work. Snapshot queries: these queries concern the

device network at a given point in time. Long-running

queries: these queries concern the device network

over a time interval.

The warehousing approach represents the state of

the art (Bonnet et al., 2000). The queries processing

on the extracted data of the sensors and the access to

the network are two distinct stages. The sensor net-

work is simply used to collect data. The query mech-

anism proceeds in two steps. First, the data are ex-

tracted of the devices of a form predeﬁned and stored

in a database server. Second, the queries process-

ing are accomplished in the server. This approach

is appropriate to answer query predeﬁned on histor-

ical data. Some disadvantages are visible, such as:

(1) waste of valuable resources for transferring great

amounts of data for the server, where many of these

data are irrelevant and (2) it is not appropriate to an-

swer snapshot and long-running queries.

Considering the disadvantages mentioned in the

warehousing approach and the increase of the sup-

ported functionalities for the sensor, a new approach

is being proposed for these networks. Called ap-

proach distributed (Bonnet and Seshadri, 2000), this

allows that some queries can be processed in the own

device and, also, that the database system to control

the resources that are used. It is primarily targeted

at snapshot and long-running queries; in addition ag-

gregate queries over historical data could be evaluated

against materialized data stored on some devices in-

stead of a centralized server (Bonnet et al., 2000). In

our work, we adopt this approach.

3 RTDB MODEL FOR SENSOR

NETWORKS

In this Section, we describe a model in Coloured Petri

nets of real-time database for sensor network. For

the modelling we use the coloured Petri nets (CPN)

(Jensen, 1997) due to the fact that they offer an envi-

ronment uniform for modelling, formal analysis and

simulation of discrete events systems, allowing to a

simultaneous visualization of its structure and behav-

ior.

In the model, we consider two sensors, two up-

date transactions, a query transaction and a server

database object. The sensors acquire the data of the

environment, store and transmit the data for the server

database object. This consequently reduce the data

ﬂow in the network and the energy waste. The server

is updated in order to allow that historical query are

accomplished, a time that is not possible to store all

the data in the sensor for a long period of time. Is

important to observe that the sensor data are always

more current than the server data. This difference

happens due to the delay that exists between an ac-

quisition of the data, for the sensor, and the update of

this data in the server.

We show the components that form the model. We

illustrate each one, considering the notation of Petri

nets, where the states (places) are represented by cir-

ICEIS 2004 - DATABASES AND INFORMATION SYSTEMS INTEGRATION

600

cles and events (transitions) by rectangles. These are

connected by arcs.

The sensor component has a place that acquires the

data (place A) and other to store them (place B). The

action of storing is represented by the rectangle shown

in Figura 1.

A

B

Acquires data of

the real world

Stores data

A’

B’

Acquires data of

the real world

Stores data

C C’

B'’

C'’

Query

Update 1 Update 2

Sensor 1 Sensor 2

Figure 1: CPN Clock Component with Transactions

The transactions and the sensors are synchronized

by a global clock, that establishes the time of the sys-

tem. The transactions and the sensors possess timing

constraints. The Figure 1 illustrates the transactions

with two places and a transition. The behavior is as

follow: the place A and A

acquires the data. Then

these data are stored in places B e B

. Periodically the

update transactions are executed, in order to keep the

server with the current data acquired by the sensors.

The query transaction, also periodically, can accom-

plish historical query in the server, as well as snapshot

and long-running queries directly in the sensors.

ID

C’

C

C'’

D

D’

Update

Query

Update 2

Update 1

Figure 2: CPN Network Component

Figure 2 shows the model of network, or either, the

way as the transactions to have access to the server.

The transactions are represented by the places C, C

. The place ID veriﬁes if it is an update or a query

transaction and the places D and D

are the places

that will have access to the database object update

or query, respectively. In the network has an inter-

mediate layer that guarantees the serializability prop-

erty during the concurrent execution of the conﬂicting

transactions. In this layer, illustrated by the rectan-

gular shadings in Figure 2, we model the negotiation

function.

It is important to observe that the model state for

the negotiation function alone is modiﬁed if the at-

tempt of concurrent execution of two distinct transac-

tions, where they are trying to have access the same

data item and at least one of them is of update.

D’

Query

DC’

BD

D

Update

DA’

Server Database

UpdateQuery

Figure 3: CPN Server Database Component

Finally, we have the referring component to the

server database object, Figure 3. This possesses two

methods, update and query. The places D and D

are the same places that the represented in the Fig-

ure that illustrates the network. In the nomenclature

of Petri nets, when a place is identical the another, in

an another page, is called of fusion place. A query

or update executed in the database, place BD, the re-

sult will be placed in the places DC

and DA, re-

spectively. These places also are fusion places of the

places D and D

, Figure 2.

4 MODEL ANALYSIS

For analysis purposes we consider a situation where

the updates and the queries are trying to access the

same data item in the database server application. We

perform both logical and timing veriﬁcation for the

data and transactions. Thus, we analyze the trans-

actions scheduling, considering the semantic concur-

rency control, where the correctness criterion is the

epsilon serializability considering quality of service.

To analyze the model we consider a situation where

two sensors acquire data of the same type. Sensor1:

writes periodically in the local database, and the re-

lease time is 1 time unit (t.u.) and the period is 6 t.u..

We consider that the value of the data item may vary

between 8 and 11 units. Sensor2: writes periodically

in the local database, and the release time is 9 t.u. and

the period is 9 t.u.. The value of the data item may

vary between 8 and 11 units. Update 1: periodically

REAL-TIME DATABASES FOR SENSOR NETWORKS

601

updates the server database object with respect to the

data in Sensor1. The release time is 3 t.u., the com-

putational time is 2 t.u., the deadline is 12 t.u. and the

period is 12 t.u.. Update 2: periodically updates the

server database object for sensor2. The release time

is 9 t.u., the computational time is 4 t.u., the deadline

is 18 t.u. and the period is 18 t.u.. Query: periodically

queries the real time server database application. The

release time is 3 t.u., the computational time is 6 t.u.,

the deadline is 12 u.t. and the period is 12 t.u..

In Figure 4, we present the message sequence dia-

grams for both the execution of transactions and sen-

sors. The Figure illustrates the execution of transac-

tions, as well as the updates in the server database.

It is also possible to observe the time properties for

each component and when the compatibility function

is evaluated due to the transactions concurrent execu-

tion. Each arrow represents the transaction execution.

In the time instant 1 a writing in Sensor1 occurred,

as indicated in the line labelled by Sensor Database.

Each execution is identiﬁed by the label Write Oper-

ation Sensor1 and the value added. In the line la-

belled Time Propr. time properties for both sensors

and transactions can be observed. Considering the

quality of service management functions as detailed

in (RibeiroNeto et al., 2003), this line represents the

speciﬁcation and monitoring functions. The ﬁrst ex-

ecution for Sensor2 is in 9 t.u., as seen in the Figure

In time 3, Query and Update 1 transactions start.

As observed in the arrows labels, two conﬂicting op-

erations will try to access a data record in the Server

Database. The ﬁrst is a query and the second an up-

date (to update the data item t1 to 8).

The transaction Query commits in time 9. The

transaction Update 1 had to wait the query and then

access the server. This can be observed by observ-

ing its computational time, where the committing oc-

curred in time instant 13, that should to be in time

instant 7. Thus, the operations could not be con-

currently executed, or either, the negotiation function

(compatibility function) was evaluated to false.

Before the transaction Update 1 commits, transac-

tion Update 2 begins execution. One more time, con-

ﬂicting operations try to have access to the server at

the same time. However, for this concurrent execu-

tion, the negotiation function was evaluated to true.

The evaluated parameters can be seen in the Figure 4.

The line Negotiation with a label with CF evaluate fol-

lowed by the parameters and its respective execution

time values. Thus, transaction Update 2 can execute

concurrently.

Time Propr Sensor Database

Write Operation Sensor1

Write Operation Sensor2

Write Operation Sensor1

Transactions

Query

Update 1

Update 2

Negotiation

CF Evaluate

Server Database

Read Operation Server

Write Operation Server

Value = 8

Sensor1[tl=1,pe=6]

Read data item t1 in the server

Up1[tl=3,tc=4,pr=12,pe=12]

Write value [8] for t1

Quer[tl=3,tc=6,pr=12,pe=12]

Value = 11

Sensor1[tl=7,pe=6]

Value = 8

Sensor2[tl=9,pe=9]

Item value t1 read in the server [0]

Write value [8] for t1

Up2[tl=9,tc=4,,pr=12,pe=12]

(New value=[8]-Current value=[8])

<=(Exp_max_Col=[10]-Exp_imp=[0])

Value = 10

Sensor1[tl=13,pe=6]

UpdateOk

Write Operation Sensor2

Write Operation Sensor1

Write Operation Sensor2

Update 1

Query

Update 2

CF Evaluate

Write Operation Server

Read Operation Server

UpdateOk

Write value [10] for t1

Quer[tl=15,tc=6,pr=12,pe=12]

Read item data t1 in the server

Up1[tl=15,tc=4,pr=12,pe=12]

[t1],Transaction Status=null

(VrUp=[10]-VrS=[8])<=(CI=[7]-II=[0])

Value = 10

Sensor2[tl=18,pe=9]

Value = 8

Sensor1[tl=19,pe=6]

UpdateOk

Item value t1 read in the server is [8]

Value = 10

Sensor1[tl=25,pe=6]

Value = 10

Sensor2[tl=27,pe=9]

Write value [10] for t1

Up2[tl=27,tc=4,,pr=12,pe=12]

Update 1

Query

CF Evaluate

Write Operation Server

Read Operation Server

Write Operation Server

Write value [10] fot t1

Quer[tl=27,tc=6,pr=12,pe=12]

(Update Value=[10]-Updatting Value=[10])

<=(Exp_max_col=[10]-Exp_imp=[0])

Read item value t1 in the server

Up1[tl=27,tc=4,pr=12,pe=12]

[t1],Transaction Status=critical

(VrUp=[10]-VrS=[10])<=(CI=[7]-II=[0])

UpdateOk

Item value t1 in the server is [10]

Figure 4: Message Sequence Diagram

ICEIS 2004 - DATABASES AND INFORMATION SYSTEMS INTEGRATION

602

5 CONCLUSION

In this paper we discussed how real-time databases

can be applied in the context of embedded systems.

We have shown the modelling and analysis, using

Coloured Petri Nets, for a real-time database server

applied in the context of a sensor network. All the

components of a sensor network system have been in-

tegrated in the model. We introduced an distributed

approach to deal with data management.

We are currently working towards the implementa-

tion of a real-time database server and its application

to a oil industry sensor network.

REFERENCES

Bonnet, P., Gehrke, J., and Seshadri, P. (2000). Query-

ing the Physical World. IEEE personal Communica-

tions. Special Issue Networking the Physical World,

7(5):10–15.

Bonnet, P. and Seshadri, P. (2000). Device Database Sys-

tems. Proceedings of the International Conference on

Data Engineering ICDE’99. San Diego, CA.

Chen, Z., Gehrke, J., and Korn, F. (2001). Query Optimiza-

tion In Compressed Database Systems. In Proceed-

ings of the 2001 ACM Sigmod International Confer-

ence on Management of Data Santa Barbara, Califor-

nia.

Hansson, J. (2002). Embedded database systems. Technical

report, Dept. of Computer Science, Linkoping Univer-

sity, and Dept. of Computer Engineering, Malardalen

University. Project is funded by CENIIT (Center

for Industrial Information Technology at Linkpings

Tekniska Hgskola) and is carried out at the Real-Time

Systems Laboratory (RTSLAB), Linkping University.

Jensen, K. (1997). Coloured petri nets-Basic Concepts,

Analysis Methods and Practical Use, volume 2.

Springer-Verlag.

Jensen, K. and al, e. (1999). ”Design/CPN” 4.0. Meta

Software Corporation and Department of Computer

Science, University of Aarhus, Denmark. On-line ver-

sion:http://www.daimi.aau.dk/designCPN/.

RibeiroNeto, P. F., Perkusich, A., and L.B.Perkusich, M.

(2003). Real-Time Database Modeling Considering

Quality of Service. 5th International Conference on

Enterprise Information Systems.

Tesanovic, A., Hansson, J., Nystrom, D., and Norstrom, C.

(2002). Embedded databases for embedded real-time

systems: A component-based approach. Technical re-

port, Dept. of Computer Science, Linkoping Univer-

sity, and Dept. of Computer Engineering, Malardalen

University.

REAL-TIME DATABASES FOR SENSOR NETWORKS

603