A DISTRIBUTED ADAPTATION MODEL FOR AMBIENT

ASSISTIVE LIVING APPLICATIONS

M. T. Segarra

IT / TELECOM Bretagne, Technopˆole Brest-Iroise, Brest Cedex 3, France

F. Andr´e

IRISA / INRIA Rennes-Bretagne Atlantique, Campus de Beaulieu, Rennes, France

Keywords:

Distributed adaptation model, Adaptive software, Ambient assistive living application.

Abstract:

Ambient assistive living applications are generally intended to allow older people to easily access medical

and emergency services, and maintain social and relatives contacts. In order to improve performance, elderly

related data may be distributed among the medical staff and/or the relatives. However, access to such data

may require different properties such as conﬁdentiality or urgency depending on the actor accessing data, the

computing environment used to access them, and their nature. Therefore, ambient assistive living applications

should be designed as adaptive software and include the necessary mechanisms to dynamically modify their

behavior. In this paper, we propose a model for dynamic adaptation that clearly separates adaptation from

business logic and that can be customized by applications designers in order to satisfy adaptation needs. The

model is based on a set of mandatory functionalities that manage basic adaptation operations and an optional

functionality that can be used to distribute adaptation mechanisms.

1 INTRODUCTION

Ambient assistive living applications are generally in-

tended to allow older people to stay at home in a

familiar environment while connected to the family

and medical staff. Such applications include sev-

eral types of services such as health care services,

communication between the elderly and their fam-

ily, information-oriented services, etc. all of them al-

lowing the concerned people to access elderly-related

data. Access to such data may require different prop-

erties depending on the nature of the data or even the

actor accessing them. The computing environment

(ﬁxed or wireless network, portable or ﬁxed work-

station...) may also inﬂuence access conditions. We

consider such applications as a set of logical envi-

ronments: the medical environment, the family mem-

bers environment, etc. Each environment may be dis-

tributed and has different adaptation needs (emergen-

cies, security, etc). Therefore, ambient assistive living

applications should be designed as adaptive software

and include the necessary mechanisms to dynamically

modify their behavior.

Dynamic adaptation has recently become an im-

portant issue when designing and developing appli-

cations. Indeed, applications may execute on differ-

ent heterogeneous environments which available re-

sources vary over time. Research efforts are mainly

focused on the proposal of basic tools for dynamic

adaptation. These tools allow applications to spec-

ify adaptation policies (interesting changes on re-

sources and required adaptation actions) and accord-

ing to them include all the necessary machinery for

deciding, planning and executing adaptation actions.

However, the proposed tools do not take into account

distribution neither of components nor of adaptation

mechanisms themselves which limits their scope.

Our current research work focus on extending

adaptation mechanisms so that adaptation designers

are able to choose a distributed architecture for their

system. In this sense, we aim at proposing a generic

adaptation model that offers a wide set of functional-

ities that can be customized by adaptation designers.

The paper is organized as follows. Nextsection in-

troduces the application we have considered in order

to propose our adaptation model. Section 3 presents

the terms we use concerning adaptation and intro-

duces the basic functionalities that should be imple-

363

T. Segarra M. and André F. (2009).

A DISTRIBUTED ADAPTATION MODEL FOR AMBIENT ASSISTIVE LIVING APPLICATIONS.

In Proceedings of the International Conference on Health Informatics, pages 363-368

DOI: 10.5220/0001774403630368

 SciTePress

mented to make a software adaptive. Section 4 ex-

tends the basic functionalities in order to add ﬂexibil-

ity when building an adaptive distributed architecture.

Finally, sections 5 and 6 position our work in the re-

search results concerning adaptation and discuss fu-

ture work, respectively.

2 TARGETED APPLICATION

The growth in population ageing in most industri-

alized nations is posing several problems at differ-

ent levels of our society (political, social, and famil-

iar/personal levels). Statistics data provide a clear pic-

ture of the problem dimensions. According to recent

results of a survey funded by the European Commis-

sion (B¨orsch-Supan and J¨urges, 2005), in 2050 the

share of the above 60 age group will be around 37 in

Europe and even more in Japan, and slightly lower in

North America (27%).

Improving the quality of life of such a population

depends on the efﬁciency, comfort and cosiness of the

place an individual calls “home”. Indeed, the preva-

lence of disability increases with age at a signiﬁcant

rate (Commission, 2007). As a result of the shifting

age proﬁle, the size of the impaired population will

increase signiﬁcantly in future years. For the elderly,

home is a place of memories where they spend a lot of

their time. Their demands on their home environment

will increase and change with growing age, especially

when their state of health starts to worsen. Yet the

ability to perform the activities of daily life with no

or little help from other, is essential to older people’s

well-being and self-esteem. And active participation

in society, through social contacts and activities, daily

economic activities such as shopping, and democratic

decision-making are key to well-being.

Therefore, an assistive living application has to of-

fer a set of services that allow older people to live

longer at the place they like most, while ensuring their

autonomy and high quality of life. We consider this

application as composed of a set of services that can

be grouped into three categories:

• health-oriented services that allow the elderly to

access to medical services, and facilitate collabo-

ration among medical staff,

• communication-oriented services that allow the

elderly to maintain social contacts,

• information-oriented services that allow the el-

derly to access information to which they are in-

terested in.

Services on different categories do not require the

same properties of the execution environment such as

security, conﬁdentiality or urgency. For example, an

alert may require a general practitioner to urgently es-

tablish a high-quality video conference contact with

the elderly in a secure manner while a less-secured,

lower-quality link may be establish between the fam-

ily and the elderly. Therefore, changes on the ex-

ecution environment (e.g. bandwidth), should lead

the application to perform adaptation actions so that

the required properties are maintained. Moreover,

urgency and preferences must be taken into account

when deciding adaptation actions. For example, an

urgent situation may need the utilization of all avail-

able bandwidth resources leading all other services to

be stopped.

On the other hand, disease(s) of an elderly person

evolve over time. Time scale of this evolution varies

from several minutes (e.g. an older person reading lo-

cal news) to several years (e.g. Alzeihmer’s disease).

An ambient assistive living application has to adapt to

the elderly capacities and dynamically change avail-

able services according to them. For example, a vocal

synthesis service may be used when the older is ex-

hausted so that she is able to continue be informed

about local news.

Therefore, adaptation of an ambient assistive liv-

ing application should cope with resources variability

in order to ensure quality of service, and elderly capa-

bilities that may vary over time. Traditionally, adap-

tation has been included in an ad-hoc manner leading

to speciﬁc development efforts and limiting the inter-

est of the provided services. Our approach is to pro-

pose a generic model for adaptation mechanisms that

can be customized to particular adaptation needs. It

is structured as a set of functionalities that allows for

monitoring execution context, deciding about adapta-

tion, planning adaptation actions and executing them.

We have been using our model for adapting central-

ized applications and we are currently considering the

ambient assistive living application and its distributed

services in order to analyze the impact of distribution

on our model.

3 ADAPTIVE SOFTWARE

ARCHITECTURE

3.1 Vocabulary

In order for a software to be adapted, one may decide

where adaptation mechanisms are associated to: the

whole application, processes, services, components,

or data. We consider each of them, that we call en-

tity, as composed of a set of, potentially distributed,

HEALTHINF 2009 - International Conference on Health Informatics

364

components that may manage data. Therefore, a site

may execute a whole entity or part of it. For exam-

ple, our ambient assistive living application may offer

a health care service that 1) manages patients medi-

cal record and 2) includes a prescription sub-service.

The whole health care service, each patient record as

well as the prescription sub-service are called entities.

Therefore, an adaptation designer may choose to as-

sociate adaptation mechanisms to the whole service,

to each patient record, to the prescription sub-service

or to all three, leading to a hierarchical organization

of adaptation mechanisms.

Adaptation system

content (e.g. patient medical record) content (e.g. prescription service)

Adaptation system Adaptation system

Adaptive primitive entity Adaptive primitive entity

Adaptive composite entity (e.g. health−care service)

Figure 1: Adaptive composite entities, adaptive primitive

entities.

An entity may include adaptation mechanisms

while an adaptive (or self-adaptive entity) should in-

clude them. As depicted in Figure 1, adaptive entities

may be primitive or composite. Adaptation mecha-

nisms of primitive adaptive entities implement adap-

tation functionalities for the entity. Adaptation mech-

anisms of composite adaptive entities coordinate the

adaptation of the included adaptive entities.

We call adaptation model the set of generic func-

tionalities and their relation that ensure adaptation of

an adaptive entity. Some of them will be manda-

tory but others will not. Specializing a functionality

leads to a particular adaptation service. Finally, the

set of adaptation services that implement adaptation

of a particular entity is called an adaptation system.

3.2 Mandatory Adaptation Model: The

DYNACO Framework

Previous work on adaptation have lead to DY-

NACO (Buisson, 2006)

, a framework that aims

at helping developers designing and implementing

A set of computers managed by the same administra-

tion.

The DYNACO framework is available at the following

web site:

http://dynaco.gforge.inria.fr/

adaptation into their own applications. The frame-

work is expressed as an assembly of abstract FRAC-

TAL (Bruneton et al., 2006) components. Develop-

ers are expected to specialize them in order to build a

framework instance speciﬁc to each application.

As Figure 2 shows, DYNACO decomposes adap-

tation mechanisms into four sub-functionalities:

• observe: DYNACO has to monitor execution

conditions; the detection of any relevant change

triggers adaptation;

• decide: given that something has changed in the

execution environment, the framework has to ﬁnd

out the best strategy the entity should use to en-

sure that QoS properties of the service are ob-

served;

• plan: once a strategy has been decided, it has to

be implemented into a plan, which orders the ac-

tions to change the behavior that will make the

adaptive entity achieve the strategy;

• execute: at last, planned actions have to be sched-

uled and executed, taking care of the synchroniza-

tion with the execution of the entity code.

Figure 2: Overview of the DYNACO component assembly.

In order to foster effective code reuse, DYNACO

encourages that those sub-functionalities are imple-

mented thanks to existing generic engines, which are

specialized speciﬁcally to the adaptive entity. DY-

NACO does not restrict to any particular engine,

and we have already experimented several of them.

For instance, the decide component has been imple-

mented as a delegation to a Java virtual machine, a

wrapper of the Jess (jes, ) rule-based expert system,

and a genetic algorithm.

Similarly, the plan component can be imple-

mented either as a straightforward mapping from

strategies to plans or as a logic-based planning algo-

rithm such as STRIPS (Fikes and Nilsson, 1971).

The execute component depends on the applica-

tion programming model. For example, we have pro-

posed AFPAC (Buisson et al., 2006) as an implemen-

tation for SPMD applications but others can be used

as well.

Finally, the observer component mainly depends

on the execution environment, rather than on the ap-

plication or the adaptation. Consequently, we con-

sider that it is provided by the execution environment.

As the framework does not depend on the adapta-

tion actions, and as the dependency to the program-

A DISTRIBUTED ADAPTATION MODEL FOR AMBIENT ASSISTIVE LIVING APPLICATIONS

365

content

Adaptation system

Adaptive primitive entity

Adaptation Coordination

Meta−Coordination

Coordination Level

Meta−Coordination Level

Adaptation Level

coordinates

2..*

coordinates

Adaptive composite entity

dialogue

Figure 3: Distributed adaptation model.

ming model of the application is captured into the

single implementation of the execute component, the

DYNACO framework is highly generic. We therefore

consider functionalities of DYNACO and their rela-

tions as our mandatory model. Each adaptive entity

should include a specialization of DYNACO in order

to implement adaptation.

4 DISTRIBUTED ADAPTATION

As mentioned in section 3.1, adaptation mechanisms

may be associated to different entity levels lead-

ing to the primitive and composite adaptive enti-

ties. When associating adaptation mechanisms to the

whole application, an adaptation designer must pro-

pose a holistic adaptation policy. This approach may

be interesting for applications that do not need much

adaptation and leads to a centralized adaptation archi-

tecture. However, for highly adaptive or distributed

applications a better approach would be to partition

the adaptation policy by considering adaptation for

each entity separately and then coordinate adaptation

performed on different entities.

Separation between adaptation and adaptation co-

ordination may be applied at different levels depend-

ing on the complexity of the adaptation policy to be

implemented. According to this, Figure 3 presents

our model for distributed adaptation. Each adaptive

primitive entity has an adaptation system associated

to it that decides, plans, and executes adaptation for it.

On the other hand, adaptive composite entities encap-

sulate an adaptation coordination system that coordi-

nates adaptation of different adaptive primitive enti-

ties and dialogues with other systems so that global

adaptation is performed. Alternatively, communica-

tion among adaptation coordination systems may be

governed by a meta-coordination system. Although

the tower for coordination may have inﬁnite layers, a

pragmatic approach may limit it to two or three.

4.1 Adaptation for Data Management

In order to validate our distributed adaptation model,

we have considered a (highly) simpliﬁed version of

our ambient assistive living application and in partic-

ular a data consistency management service (CMS).

This service may be used by medical staff in the same

or different hospitals, mobile or not, to maintain con-

sistency among replicas of patients’ records. De-

pending on the execution context

, different consis-

tency protocols may be used. For example, when high

network bandwidth is available, a strong consistency

protocol is well suited as it ensures that all replicas are

consistent anytime and the user will not observe de-

lays on reading or writing. On the other hand, when

low network bandwidth is available, a weak consis-

tency protocol is better suited as it allows replicas to

be accessed even if network partitions exist.

For scalability and fault tolerance reasons, the

CMS has been implemented as a distributed service.

An update entity is associated to each replica that al-

lows accessing (reading and writing) the local replica

and notifying other replicas about local updates. The

adaptation policy to be implemented advocates the

use of the strong consistency protocol when possible

(high bandwidth availability).

Traditionally, consistency management services

provide a single protocol to manage different repli-

cas. Indeed, they assume the environment

as a whole

and propose the protocol that better suits the context

offered by it, i.e. the “worst case”. For example,

even if several replicas are well-connected, the Bayou

replicated database will use a weak consistency pro-

tocol for all replicas if one of them is weakly con-

nected (Terry et al., 1998). In order to avoid such a

problem, we consider the environment as a set of sub-

environments each of which 1) offers a context, and

2) does not overlap other sub-environments, i.e. de-

vices on two sub-environments do not share network

connections with the same properties.

Figure 4 shows our CMS and the adaptation archi-

tecture associated to it. For simplicity reasons, the ﬁg-

ure shows replicas (four) of one particular data. The

environment is divided into two sub-environments.

Available memory, CPU, bandwidth, etc.

Set of devices and network connections that use the ser-

vice.

HEALTHINF 2009 - International Conference on Health Informatics

366

Adaptation System

Weak consistency

Adaptive update entity

Adaptation System

Weak consistency

Adaptive update entity

System

Adaptation Coordination

Adaptation System

Strong consistency

Adaptive update entity

Adaptation System

Strong consistency

Adaptive update entity

System

Adaptation Coordination

coordinate

dialogue

Sub−environment A

Sub−environment B

Figure 4: Distributed adaptation architecture for consis-

tency management.

Sub-environment A offers high network bandwidth;

replicas on it are managed using a strong consis-

tency protocol. On the other hand, replicas on sub-

environment B are managed by a weak consistency

protocol as bandwidth availability is low.

As shown in the ﬁgure, the adaptation architec-

ture is two-layered. Each update entity has an adapta-

tion system associated to it that changes from strong

to weak consistency and vice-versa on bandwidth

availability changes. Coordination is performed at

the sub-environment level; adaptation on each sub-

environment is managed by an adaptation coordina-

tion system that dialogues with its counterpart on the

other environment in order to perform global adapta-

tion. The architecture respects our adaptation model.

Each update entity is modeled as an adaptive primi-

tive entity. Update entities on an environment are en-

capsulated in an adaptive composite entity and their

adaptation is coordinated by the Adaptation Coordi-

nation System.

5 RELATED WORK

Several architectures and frameworks have been pre-

viously proposed to implement dynamic adaptation.

These architectures usually emphasize the require-

ment for reﬂective programming support in order to

implement the adaptation actions, such as in (Amano

and Watanabe, 1999), (Keeney and Cahil, 2003),

and (Capra et al., 2001). Indeed, reﬂexivity provides

means to modify application behavior, in the underly-

ing runtime environment or in the adaptation frame-

work itself. In our approach, the need for reﬂexivity is

captured within adaptation actions. In fact, our model

does not specify how actions are implemented; never-

theless, parametrability, reﬂexivity and runtime code

modiﬁcation are the major tools. Moreover, neither

of these works takes into account distribution of the

adaptation mechanisms which limits their interest.

Other approaches such as (Lapayre and Renard,

2005) and (Vadhiyar and Dongarra, 2005) are based

on the runtime environment and/or compilers to make

applications adaptive. Therefore, the set of possible

adaptation strategies is restricted by the developers

of the runtime environment and/or compilers. In our

approach, the adaptation developer is responsible for

customizing the framework according to application

needs.

In (Yan and Sere, 2004) formalizations of dy-

namic adaptation are proposed to verify correctness

of adaptation. However, they do not provide help in

designing the adaptation itself. This approach is com-

plementary to ours, which focuses on helps for de-

signing the adaptation while dismissing the problem

of the veriﬁcation.

Only few studies have been conducted about the

methodology of dynamic adaptation. In (McIlhagga

et al., 1998), authors list relevant questions that de-

velopers should answer to design adaptive software.

However, this work does not provide a clear structure

of an adaptive application, potentially leading to ad-

hoc implementations and poor reusability. Moreover,

distribution of adaptation is not taken into account.

6 CONCLUSIONS AND FUTURE

WORK

In this paper we have presented a model for structur-

ing dynamic adaptation mechanisms intended to be

A DISTRIBUTED ADAPTATION MODEL FOR AMBIENT ASSISTIVE LIVING APPLICATIONS

367

used in a distributed environment and its utilization

for a consistency service used in an ambient assistive

living application. The model assumes applications as

a set of “entities” (components, services) which may

include adaptation mechanisms as a set of functional-

ities. Some of these functionalities are mandatory and

relate to decide, plan and execute adaptation actions;

others are optional and relate to coordinate adaptation

actions on different entities. Indeed, depending on the

complexity of the adaptation policy, designers are free

to partition the policy so that several entities encapsu-

late their own one. Coordination may be necessary to

ensure a consistent global adaptation.

We have already implemented and tested a cen-

tralized version of our model, without coordination

mechanisms (Buisson et al., 2006). Ongoing work

concerns the implementation of a ﬁrst prototype of

our model applied to the consistency service as de-

scribed in section 4. The prototype is based on the

Fractal component model and its Java implementa-

tion Julia (Jul, ) as its hierarchical model ﬁts well

into our primitive/composite entities. The prototype

will allow us to better understand how to propagate

and coordinate changes through the system, evaluate

costs and beneﬁts of distribution, and to precise how

to structure an environment (or an adaptation frame-

work) in face of overall dynamic properties. More-

over, new functionalities may be identiﬁed and the

coordination/meta-coordination should be reﬁned in

order to clearly separate decision, planning, and exe-

cution coordination.

REFERENCES

Jess, the rule engine for the java platform.

herzberg.ca.sandia.gov/.

Julia home page. fractal.objectweb.org/julia/index.html.

Amano, N. and Watanabe, T. (1999). An approach for

constructing dynamically adaptable component-based

software systems using lead++. In OOPSLA’99 Inter-

national Workshop on Re ection and Software Engi-

neering.

B¨orsch-Supan, A. and J¨urges, H. (2005). The survey of

health, ageing and retirement in europe - methodol-

ogy. Technical report, Mannheim Research Institute

for the Economics of Aging.

Bruneton, E., Coupaye, T., Leclercq, M., Qu´ema, V., and

St´ephani, J.-B. (2006). The fractal component model

and its support in java. Software Practice and Ex-

perience, , special issue on Experiences with Auto-

adaptive and Reconﬁgurable Systems.

Buisson, J. (2006). Adaptation dynamique de programmes

et composants parall`eles. PhD thesis, INSA de

Rennes, IRISA.

Buisson, J., Andr´e, F., and Pazat, J. (2006). Afpac: En-

forcing consistency during the adaptation of a parallel

component. Scalable Computing: Practice and Expe-

rience.

Capra, L., Emmerich, W., and Mascolo, C. (2001). Re-

ﬂective middleware solutions for context-aware appli-

cations. In Third International Conference on Met-

alevel Architectures and Separation of Crosscutting

Concerns.

Commission, E. (2007). Ageing well in the information so-

ciety - an i2010 initiative - action plan on information

and communication technologies and ageing. Techni-

cal report, European Commission.

Fikes, R. and Nilsson, N. (1971). Strips: A new approach

to the application of theorem proving. Artiﬁcial Intel-

ligency, 2.

Keeney, J. and Cahil, V. (2003). Chisel: a policy-driven,

context-aware, dynamic adaptation framework. In 4th

International Workshop on Policies for Distributed

Systems and Networks.

Lapayre, J. and Renard, F. (2005). Appat: A new platform

to perform global adaptation. In International Confer-

ence on Distributed Frameworks for Multimedia Ap-

plications.

McIlhagga, M., Light, A., and Wakeman, I. (1998). To-

wards a design methodology for adaptive applications.

Mobile Computing and Networking.

Terry, D., Petersen, K., Spreitzer, M., and Theimer, M.

(1998). The case for non-transparent replication: Ex-

amples from bayou. IEEE Data Engineering.

Vadhiyar, S. and Dongarra, J. (2005). Self adaptability

in grid computing. Concurrency and Computation:

Practice and Experience.

Yan, L. and Sere, K. (2004). Formalism for context-aware

mobile computing. In Third International Workshop

on Algorithms, Models and Tools for Parallel Com-

puting on Heterogeneous Networks.

HEALTHINF 2009 - International Conference on Health Informatics

368