Section: New Results

Dynamic Adaptation of Service-based Applications

Participants : Françoise André, Djawida Dib, Erwan Daubert, Guillaume Gauvrit, André Lage, Christine Morin, Nikos Parlavantzas, Jean-Louis Pazat, Chen Wang, Mohamed Zouari.

Dynamic Adaptation in a Distributed Operating System

Participants : Françoise André, Djawida Dib, Christine Morin, Nikos Parlavantzas.

We have studied the feasibility to dynamic adapt the features of a distributed operating system using a framework for self-adaptation of service oriented distributed applications [46] . We have focused on the consistency protocols for replicated data in distributed shared memory systems. We have considered two strict consistency protocols, one based on invalidation and one based on broadcast on write operations. The adaptation framework selects one of these two algorithms based on the inter-node data transfer delay. We have implemented a prototype based on Kerrighed single system image operating system for clusters and the SAFDIS adaptation framework. We have integrated a broadcast based consistency protocol in Kerrighed that already implements a write invalidation consistency protocol. We have implemented the adaptation policy in the SAFDIS framework and the needed adaptation mechanisms in Kerrighed as well as a component for monitoring data transmission delays. An experimental evaluation is being carried out.

Adaptation for Data Management

Participants : Françoise André, Mohamed Zouari.

The usage of context-aware data management in mobile environments has been previously investigated by Françoise André in collaboration with Mayté Segarra and Jean-Marie Gilliot from Telecom Bretagne Brest (previously known as ENST Bretagne). This work focuses on data management in grid and mobile environments; an ambient assisted living application illustrates the approach. This work was realized in the context of the ALORAD project (Architecture LOgicie lle pour la Réplication Adaptative de Données), financed by the Brittany council. Mayté Segarra from Telecom Bretagne Brest was co-adviser for the PhD thesis of M. Zouari [12] .

Adaptation for Service-Oriented Architectures

Participants : Françoise André, Erwan Daubert, Guillaume Gauvrit, André Lage, Nikos Parlavantzas, Jean-Louis Pazat, Chen Wang.

Service-Oriented Computing is a paradigm that is rapidly spreading in all application domains and all environments - grids, clusters of computers, mobile and pervasive platforms. The following works take place in the context of the S-Cube European Network of Excellence.

Services adaptation in distributed and heterogeneous systems

We are still studying service adaptation in distributed and heterogeneous systems. This work covers different aspects such as structural, behavioral and environmental adaptation, distributed decision and planification of adaptation actions, adaptive allocation of resources for services. A framework called SAFDIS for "Self Adaptation For Distributed Services" has been defined and implemented. It is built as a set of services, providing functionalities useful to build an adaptation system. The analysis phase can take reactive as well as proactive decisions. This gives the ability to either react fast or to take decisions for the long term. This implies the ability to analyze the context with a variable depth of reasoning. Our implementation of the SAFDIS analysis phase also distributes and decentralizes its analysis process to spread the computational load and make the analysis process scalable. The planning phase seeks the set of actions (the plan) needed to adapt the system according to the strategy chosen by the analysis phase. It also schedules the selected actions to ensure a coherent and efficient execution of the adaptation. The planning topic is a well known subject in AI research works and many algorithms already exist in that field to produce efficient schedules. With our SAFDIS framework, the planning phase is able to reuse these algorithms. The resulting plan of actions can have actions that can be executed in parallel.

Quality Assurance for Distributed Services

In the context of the service-centric paradigm, we have designed and developed the Qu4DS (Quality Assurance for Distributed Services) research prototype. Qu4DS is a cloud PaaS solution which fills the gap between the conception of higher-level SaaS service providers over the resource-level PaaS layer. Qu4DS provides an automatic support for service execution management by aiming at increasing the service provider's profit. More specific, Qu4DS dynamically acquires resources according to the customer demand, deploys service instances and implements QoS assurance mechanisms in order to prevent SLA violations. Moreover, Qu4DS has been evaluated on Grid5000 and showed to be effective on reducing service provider's costs [27] .

Self-configuration for Cloud Platforms

By definition, cloud computing offers an abstraction to manage various needs and concepts such as distributed software design, the deployment of such software on dynamic resources and the management of this kind of resources. Thus it is possible to reconfigure (adapt) according to some needs the software as well as the use of the resources. However these reconfigurations that are used on different layers may also have impacts on the others. Moreover these layers are independent and so are able to adapt themselves independently of the others. In our work, we propose to use some adaptation capabilities offered for example by the infrastructure (IaaS) that manages the resources to adapt the software (SaaS). We also propose to use planning algorithms to coordinate the adaptations between them to avoid inconsistency or inefficiency due to concurrent adaptations.

Dynamic Adaptation of Chemical services

We have proposed a QoS-aware middleware for dynamic service execution. In the context of dynamic execution, a workflow is defined by composing a set of abstract activities as place holders. Each activity is bound to a suitable partner service, which is selected at run-time from a set of functional equivalent candidates with different non-functional properties such as quality of service (QoS). The service selection process is modeled as a series of chemical reactions.

Prediction of SLA violations and dynamic adaptation in workflows

During execution, run-time QoS is determined by the dynamic execution environment and thus the expected QoS is not always ensured. In addition, infrastructure failures can make a service undeliverable. The adaptive execution reflects the capability to recompose a (part of a) workflow on-the-fly in case that global SLA violation is predicted. Most techniques for predicting global SLA violation require past experiences on executions of a business process. All historical execution instances have the same structure as well as the same bindings. These solutions do not adapt to the case of dynamic execution, where for each execution, partner services are selected and bound at run time.

In order to predict global SLA violation in the context of dynamic service execution, we proposed a 2-phase prediction technique, which is fit for generic workflow composition. The prediction method works with a high accuracy for simple workflows, but when the workflow composes complicated structures (such as loops and exclusive branches), the performance degrades. The reason is that the estimation of global SLA is based on the critical path, which is not definitely executed. To solve this problem, we propose to use data mining technique to predict workflow branches and the number of loop execution. Based on predicted branches, the prediction of global SLA violation is much more accurate. The numerical evaluation will be carried out in the near future.