The ARLES project-team investigates solutions in the forms of languages, methods, tools and supporting middleware, to assist the development of eistributed software systems, with a special emphasis on mobile distributed systems enabling the ambient intelligence/pervasive computing vision. We in particular study ambient intelligence system architectures based on the service paradigm, from service modeling to service-oriented middleware for advanced wireless networks. Our research activities over the year 2008 have focused on the following complementary areas:

With network connectivity being embedded in most computing devices, networking environments are now pervasive. As a result, any networked device may not only seamlessly consume but also provide software applications over the network. Service-Oriented Computing (SOC) then introduces natural design abstractions to deal with pervasive networking environments. Indeed, networked software applications may conveniently be abstracted as autonomous loosely coupled services, which may be combined to accomplish complex tasks. In addition, the concrete instantiation of SOC paradigms provided by Web Services (WS) technologies by means of Web-based/XML-based open standards (e.g. WSDL, UDDI, HTTP, SOAP) may be exploited for concrete implementation of pervasive services. However, while Web services standards and implementations targeting wide-area domains are effective technologies, supporting Web service access in pervasive networking environments is still challenging. In such kind of networking environments, mobile applications, acting as both service consumers and providers, often run on scarce resource platforms such as personal digital assistants and mobile phones, which have limited CPU power, memory, and battery life. Moreover, these devices are usually interconnected through one or more heterogeneous wireless links, which compared to wired networks are characterized by lower bandwidths, higher error rates, and frequent disconnections.

A key feature of pervasive networking environments is the diversity of radio links available on portable devices, which may be exploited towards seamless connectivity. Specifically, as nodes get connected via multiple radio links, thorough scheduling and handover across those links allows enhancing overall connectivity and actually making it seamless. This calls for making services network-agnostic, so that the underlying middleware takes care of scheduling exchanged messages over the embedded links in a way that best matches Quality of Service (QoS) requirements, and further ensures service continuity through vertical handover. In this setting, a primary requirement for supporting service-oriented middleware is to provide a comprehensive networking abstraction that allows applications to be unaware of the actual underlying networks while still exploiting their diversities in terms of both functional and extra-functional properties.

To address the above, we have introduced the ubi SOAPcommunication middleware  , which underlies SOAP-based middleware and strives to provide pervasive networking to services. As sketched below, the architecture of ubi SOAPlayers the following constituents below SOAP-based middleware functionalities: (i) multi-radio networking provides network-agnostic connectivity, (ii) multi-network routing implements a multi-network overlay, and (iii) SOAP Communication leverages multi-radio, multi-network message routing, and further introduces communication primitives targeted at pervasive computing systems. As discussed in Section  , the prototype implementation of ubi SOAPis released under open source license.

Multi-Radio Networks.The multi-radio networking layerof ubi SOAPprovides core functionalities to effectively manage multi-radio connectivity by providing a network-agnostic addressing scheme together with QoS-aware network link selection. In particular, the multi-radio networking layeris in charge of:

Multi-network Overlay.Thanks to the ubi SOAPmulti-radio networking layer, communication among nodes exploits the various network links that the nodes have in common, further selecting the link that provides the required QoS. However, in some cases, it might also be desirable for nodes to be able to access services that are hosted in networks to which the requesting node is not directly connected (e.g., to provide continuity of service despite node mobility). For this purpose, ubi SOAPintroduces an overlay network that bridges heterogeneous networks, thus enhancing overall service connectivity. Specifically, nodes that are connected to two (or more) different networks through their network interfaces can assume the role of bridge nodes. Bridge nodes quite literally “bridge” between two separate networks, relaying point-to-point and multicast ubi SOAPmessages across those networks. In particular the multi-network routinglayer is in charge of:

Custom SOAP Transport for Pervasive Networking.In order to leverage the provided multi-radio, multi-network service connectivity, ubi SOAPintroduces custom SOAP point-to-point communicationthat brings multi-radio multi-network routing to legacy SOAP messaging. Further, ubi SOAPimplements a SOAP transport for group communicationthat enhances the SOAP API to meet the corresponding base requirement of pervasive networking environments. In particular, the SOAP Communicationlayer is in charge of:

Service Discovery over ubiSOAP.To validate the above, we have realized a Pervasive Service Discovery (PSD) Service that provides dynamic, interoperable, context-aware service discovery by benefiting from all the advanced features of ubi SOAP, i.e., group communication, and multi-radio, multi-network message routing. PSD is mainly a reengineering of the open source MUSDAC multi-protocol service discovery platform (see §  ) on top of ubi SOAPin order to support service discovery in multi-radio, multi-network environments.

PSD uses a hierarchical approach for service discovery in multi-network environments. Indeed, a (logically) centralized repository (PSD-S) coordinates service discovery within an independent network, while PSD-Ss in different networks communicate together in a fully distributed way to disseminate service information. While in MUSDAC service discovery and access were tightly integrated, PSD-Ss are only concerned with service discovery, and rely on ubi SOAPgroup communication to disseminate service information across networks. Changes in the multi-network topology (e.g., broken propagation paths) are then taken care of transparently.

PSD-Ss provide an explicit API supported by PSD pluginsthat enables clients (resp. providers) in a network to discover (resp. advertise) a service in the multi-network environment. It further enables clients and providers to benefit from advanced discovery features (e.g., context-awareness) by directly issuing requests or advertisements in the PSDL format. Specific legacy SDP pluginsregister with the active SDPs in the network, and translate requests and advertisements in legacy formats to PSDL (e.g., SLP and UPnP). PSD combines various matching algorithms to support the various elements of the service description (for both requests and advertisements), and thus provides comprehensive interoperability between SDPs. Finally, PSD controls the dissemination of local requests and the compilation of the results returned by distant PSD-Ss.

As described above, the PSD repository stores PSDL service descriptions that are either generated by legacy SDP plugins (e.g., UPnP2PSD plugin) or directly registered by service providers using the PSD plugin. We use a hierarchical service description format that actually combines a number of distinct documents specifying different facets of the service. The PSDL description acts primarily as a top-level container for additional files describing facets of the service. For example, a WSDL document may be used to describe the service interface while non-functional properties can be described using existing QoS and context models. The ubi SOAPgrounding of host, that identifies the networks and IP address at which the service's host is network-reachable (i.e., MRN@ and mapping) is described in such separate document thus facilitating dynamic updates.

As discussed in the previous section, service-oriented middleware (SOM) appears to be most appropriate to tackle the requirements of pervasive environments by abstracting the software and hardware resources of such environments as services and by supporting service discovery, access and composition. Nevertheless, the affluence of SOM technologies and platforms that have been put forward to address the heterogeneity and dynamics of pervasive environments has engendered a new kind of heterogeneity, i.e., middleware heterogeneity. This heterogeneity concerns the protocols associated with base middleware functionalities, which are service discovery and service access, as well as the multitude of networks in which service providers and requesters may reside. Thus, a SOM for pervasive computing should provide multi-protocol and multi-network interoperability mechanisms (see the previous section). Still, even after interoperability has been established at the networking and middleware levels, the dynamic discovery and composition of networked services by applications further require service providers and requesters to agree on the semantics of services, so that these can integrate and interact in a way that guarantees dependable service provisioning and consumption. Such an agreement cannot be carried out at the syntactic level, i.e., by assuming that service providers and requesters use a common syntax for denoting service semantics, as this is too strong an assumption for pervasive environments. Then, a promising approach towards addressing syntactic heterogeneity relies on semantic modeling of service features by employing relevant semantic technologies. Nevertheless, assessing the conformance between service semantics as announced by service providers and requested by service requesters induces costly semantic reasoning (in terms of time and computation), which makes existing solutions inappropriate for the highly interactive and resource constrained pervasive environment. Finally, besides dealing with the functional features of services, user-centrism of pervasive environments calls for the awareness of service non-functional features, i.e., Quality of Service (QoS). To address the above challenges, we have introduced an efficient, semantic, QoS-aware service-oriented middleware for pervasive computing which integrates advanced semantic service model for pervasive services, efficient and lightweight semantic service registry  , and automatic service composition  . The resulting COCOA service-oriented middleware is further released under open source license (see §  ).

Looking further into the dynamics of pervasive services, we have studied the issue of dependability. Runtime service reconfiguration is put forward as one of the means by which we could provide dependable service-oriented architectures (SOA) that enable continuity in service provisioning, and are flexible and robust in the presence of change. Indeed, in pervasive computing environments networked entities join and leave the environment at their convenience without prior notification. This dynamics comes at the price of dependability, due to runtime variations in terms of (i) service availability, and (ii) network connection/infrastructure availability, according to user/service mobility.

In this context, the main focus of our work is to incorporate support for runtime reconfiguration in SOA systems in order to tolerate runtime variations and ensure continuity in service provisioning for the users. In particular, we focus on middleware support for runtime service reconfiguration, since compared to application-level or network-level approaches, middleware seems to provide the required level of abstraction and genericity to deal with the challenges posed. Our main contribution consists in enabling service continuity by (i)  substitutinga service that becomes unavailable at runtime with a semantically similar one, and (ii)  transferringand translatingthe current state of interaction to the substitute service in order to resume the execution from the point it was interrupted. The need for state translation is due to the environments' heterogeneity, since the unavailable and substitute services are not assumed to be identically implementedand/or described.

The runtime reconfiguration thus includes automatically substituting an unavailable service and transferring its state in a way that is compatible with the substitute service. Ensuring compatibility includes:

In the case where the unavailable sevice is part of an orchestrated service compition, the impact of substitution on the still available services of the composition should also be considered. Indeed, depending on the point where the orchestration execution was interrupted and the point of execution where state transfer is possible between the unavailable and the substitute service, a roll-back may be needed for the substitute service with respect to execution point where the unavailable service was interrupted. Then, due to data dependencies, some of the still available services of the composition may also have to roll back. To deal with this, we establish checkpointing and introduce algorithms for rollback computation.

The outcome of our contribution is SIROCCO (ServIce Reconfiguration upOn serviCe unavailability and Connectivity lOss), a middleware infrastructure that enables transparent runtime reconfiguration of SOA systems upon services unavailability. The middleware discovers candidate substitute services, which are semantically compatible services that can be used in place of the service that becomes unavailable, identifies the best amongst these candidates and performs the substitution. In the case of service composition, the middleware also takes into account the impact on the still available services of the composition.

Automatic service compositionis one of the most challenging issues in Service-Oriented Computing (SOC), as it enables to provide the end-users with functionalities composed out of services existing in their environment, or more generally to have services collaborating in added-value composite services. Automatic service composition techniques rely on four interface description levels: signature (operation profiles), behavior (protocols), non functional (time, QoS) and semantics. However, these techniques usually assume that services have been previously developed to be integrated, and the proposed composition processes are limited to simple correspondences between service functionalities.

Software adaptationhas provided solutions for component interoperability through the computation – from component interfaces and user-defined adaptation abstract specifications, called mappings – of adaptors that operate in-between components to ensure their correct composition at the signature and behavioral levels. In  we have proposed a state-of-the art behavioral level adaptation technique based on transition systems and Petri net encodings to solve mismatch between components. Further, different studies have been undertaken to make adaptation applicable to Service-Oriented Architectures (SOA). This followed three complementary directions: adaptation of service orchestrations (centralized service compositions), adaptation of service choreographies (distributed service compositions), and integration of adaptation features in service composition processes. In all cases, we have developed prototypes to validate our approaches.

Efficient Behavioral Adaptation for Service Orchestration.Adaptation processes are either restrictive (pruning interaction sequences leading to deadlocks) or generative (enabling message reordering and generation when required). In our previous works we have developed a restrictive and generative adaptation approach  . Still, efficiency was an issue as pruning and adaptor behavioral reduction was performed after a raw (big) model had been generated. In  we have defined a more efficient adaptation technique, where both pruning and reduction are performed whilethe adaptor is computed. This approach is based on the encoding of service protocols (conversations) and value-passing adaptation contracts in the LOTOS process algebra and the use of on-the-fly model-checking and behavioral reduction techniques. Further, application to SOA has also been tackled by the automatic transformation of adaptor models into service orchestrations, which is achieved in two steps: (i) adaptor model filtering (to remove non-implementable message sequences) and, (ii) BPEL encoding.

Behavioral and Semantic Adaptation for Service Choreographies.While automatic adaptation is highly desirable for SOA where systems are composed from dynamically discovered services, component adaptation techniques do not support the semantic level and therefore require a mapping to be given by a designer to deal for example with message name mismatch between services. Service composition is usually achieved through orchestration. Still, distributed adaptation is an important issue in domains such as pervasive computing, due to the use of resource-limited devices and ad hoc networks, no centralized server being available to execute the orchestration/adaptor. In  , we have tackled composition and adaptation from a complementary point of view, choreography. With reference to earlier works, adaptation is fully automatic thanks to an ontology that is used to generate implicit links between the choreography partners, which no longer requires an adaptation mapping to be given beforehand. Another interest of this approach is that it generates a distributed set of adaptors and avoids centralized adaptor implementations. More recently, we have studied this approach to overcome mismatch between client protocols and roles in auction protocol choreographies.

Adaptation Features in the Service Composition Process.Task-Oriented Computing (TOC) envisions a user-friendly world where user tasks would be achieved by the automatic assembly of resources available in the environment. Service-Oriented Computing (SOC) is a cornerstone towards the realization of this vision, through the abstraction of heterogeneous resources as semantic services described in terms of capacities and data flows, that is more abstractly than in our previous works based on protocols over operation calls. Still, this description level raises both vertical (between user task and service descriptions) and horizontal (between service data flows) mismatch issues. In  we have investigated a task-oriented service composition process based on graph planning techniques (graphplan) with two specific adaptation features: vertical mismatch is supported using task decomposition guidelines, horizontal mismatch is supported using ontologies. This technique generates service compositions (orchestrations) under the form of plans which are then translated into the BPEL service orchestration language.

Ongoing work is threefold. A first perspective is related to the development of on-the-flyalgorithms for the behavioral reduction technique we have developed in 2007, and for the BPEL adaptor filtering presented in  . The objective is to gain in efficiency by applying both algorithms whilecomputing the adaptors (either their centralized or decentralized forms) rather than after a raw model has been generated. A second perspective is related to the implementation of the distributed adaptors  , using state-of-the art exogenous coordination languages. As a longer term perspective, we plan to study the application of automatic theorem provers and SAT-solvers to extend the adaptive planning technique  to services whose semantics would be described with pre- and post-condition based on first-order logic.

Privacy is a major concern in today's information society. As computers are used to mediate a growing number of human activities, a greater amount of personal data is digitalized and can be easily transmitted, stored and analyzed. If, in the past, private information such as shopping habits, activities, preferences and whereabouts vanished with time, today this data can be either directly collected or deduced from other available data, and correlated to infer a number of an individual's personal details without one's awareness. Privacy protection becomes even more significant in pervasive computing systems, where the surrounding space incorporates computing devices that the user manipulates unconscious of their existence to execute daily activities. Pervasive information systems, thus, can collect a larger amount of personal data and amplify the possibilities for privacy invasions.

To effectively protect user's privacy, the infrastructure, middleware and application layers must cooperate, supported by a privacy legislation that enables legal measures against identified privacy invasions. In our work, we focus on the pervasive middleware, which must not only provide mechanisms to encourage the development of privacy-aware pervasive applications but also must be designed taking privacy into account. Particularly, we consider privacy issues that arise on service-oriented middleware both when applications interact with middleware services as well as on the middleware design and implementation. Our work focuses on the privacy risks introduced by the three basic functionalities of a service-oriented middleware for pervasive computing, namely service discovery, service compositionand service access.

Regarding service discovery, we had already introduced and demonstrated a privacy-enhanced protocol for syntactic and semantic service discovery that reduces the trust requirements of service directories while enhancing privacy protection for both clients and service providers. Service descriptions are modified in such a way that a privacy-enhanced description can represent multiple original descriptions. Service directories are unable to relate a given privacy-enhanced service request or advertisement to the original request or advertisement, but are still able to find matches between privacy-enhanced service descriptions. As a result, service directories cannot recognize which services are announced or requested, which reduces the privacy risks posed by untrusted directories. Still concerning service discovery, we also had studied privacy issues that appear when performing multi-network service discovery. We proposed a series of mechanisms to (i) allow clients to control whereservice requests are forwarded, (ii) allow clients to control howservice requests are forwarded and (iii) enhance middleware privacy-awareness when forwarding service discovery request and results to untrusted networks.

In 2008 we extended our work to handle privacy issues in service composition and service access. When composing services in open environments such as pervasive computing, hardly ever the is exact service description that the client requests available in the environment. To enable users to execute composite services in such scenarios, the pervasive middleware must provide a flexible mechanism for service composition that can compose available services at run-time to provide a composition functionally equivalent to the user-defined composite service. This feature, however, has an impact on the user privacy since the task partition specified on the user-defined abstract workflow may not be respected by the middleware-provided executable workflows, and service providers may be able to correlate data that the user wishes to keep disconnected. In  we propose a model based on an extension of Fuzzy Cognitive Maps that enable users to reason about the privacy impact of executing a service composition and to compare the consequences on privacy of running two functionally equivalent compositions featuring distinct partitions of tasks among service providers.

Finally, we addressed the issue of service access in hybrid mobile ad hoc networks in  . Pervasive computing environments combine structured and ad hoc networks using different technologies (WiFi, Bluetooth, Ethernet, GPRS, etc.). In such scenarios, it is common that packets are routed through untrusted user nodes and proper care must be taken to avoid that unauthorized nodes have access to message contents. Even though techniques based on public key encryption to solve that problem already exist and are widely deployed in environments such as the Internet, they are not suitable to ad hoc networks that cannot rely on an infrastructure for public key distribution. We propose to explore the diversity of paths that may exist between any two nodes in such networks to improve resistance to eavesdropping. Nodes can split a message into multiple parts and send each part through a different path in such a way that to have access to the whole message an attacker must possess a number of shares, making eavesdrop attacks harder to perform. The novelty of our approach is that we consider each node's software platform and each network's characteristics to compute paths that are resistant to attack to a given software layer (e.g. an operating system) or network technology (e.g. the encryption algorithm used by a network standard).

Nowadays, users rely on various sets of connected devices and services to interact with each other: home gateways, laptops, smartphones, UMPC, enterprise servers, Web/cloud computing services, etc. While for a long time interactions have been mostly Web-centric, users now routinely engage in user centric interactions such as content sharing and social networking through mobile devices. Still, while mobile devices have become very capable of communicating with other devices, accessing or exchanging content is far from being easy: communication costs are still high, communication protocols are not uniformly supported, communication links are not always reliable, etc. Another problem lies in the role usually assigned to mobile devices. Although mobile devices are the primary terminals for interaction, they have essentially focused on (i) acting as clients accessing services in the infrastructure (e.g., enablers in the IMS infrastructure for Telecoms networks, or Web services in the Internet), or (ii) manipulating content stored locally (e.g., multimedia content playback). So far, there has been little success in truly integrating these mobile devices in the pervasive computing environment (i.e., acting also as resource or service providers). Furthermore, the use of multiple devices, having large storage capabilities, has resulted in the scattering of content on a set of devices, which, ironically, is making access to all the content that one possesses a big challenge (Where is this file? Is it the latest version?). This problem becomes acute when it comes to impromptu collaboration (e.g., in business meetings, conferences, or on the road). Such collaborations often require exchange of content, and have to wait till the user gets physical access to the device over which the required content (or the correct/latest version) is currently stored. Delay in accessing the content often cause frustration and missed opportunities.

To answer these challenges, and better support interactions between mobile users, we are developing the iBICOOP middleware  . Our middleware addresses these challenges by targeting both fixed and mobile devices, leveraging their characteristics (e.g., always on and unlimited storage for home/enterprise servers, ad hoc communication link between mobile devices), and by leveraging the capabilities of all available networks (e.g., ad hoc networks, Internet, Telecoms infrastructure networks). It also relies on Web and Telecoms standards to promote interoperability.

The base architecture of the iBICOOP middleware consists of core modules on top of which we can develop applications that may arise in up-coming multi-device, multi-user world.

We are currently implementing the iBICOOP middleware in Java, using IBM J9's JVM with CDC 1.1 for Windows mobile PDA and smartphones, as well as native CLDC/MIDP for other phones and smartphones. On the infrastructure side, services (e.g., Communication proxy service and Discovery service) are developed in Java 1.6, and deployed on Apache Tomcat and on the Alcatel 5350 IMS application server (for the IMS infrastructure). A number of core modules have already been implemented (Partnership Manager, Security Manager, and Communication Manager) on specific devices and are being ported. We plan to implement iBICOOP as a tight integration between services on the Internet, local networks, and in the IMS architecture thus providing a solution for both the Telecoms and Internet world.

We are also building two collaboration services on iBICOOP core modules: the Replication Serviceand the Exchange & Sharing service.

In iBICOOP, we are developing a middleware to allow ubiquitous access of user data from a multitude of devices with heterogeneous capabilities and running on different platforms. The services that we are building on top of iBICOOP aim to show the viability of iBICOOP as a standard-based platform for future advance services. Leveraging Telecoms and Internet world, integration with IMS architecture, is one of the salient feature of iBICOOP. The iBICOOP middleware offers a complete solution for end-user with regards to content-sharing and data management in emerging pervasive computing environments.