Cubik: Ubiquitous distributed components

Participants

• Yves Mahéo, assistant professor
• Didier Hoareau, PhD student

Context and objectives

With the emergence of more and more computing devices equipped with wireless comunication capabilities, new target platforms for distributed applications have to be considered. These platforms may still include powerful and robust machines but they are rather composed of resource-contrained and mobile devices (laptops, personal digital assistants, smartphones, sensors, etc.).

In parallel, distributed application tend to be so complex that efficient software engineering methods have to be employed in order to handle their whole lifecycle. The component-based approach has proved to be useful for ths purpose.

Project Cubik intends to contribute to the study of the component-based approach for the deployment and the execution of distributed applications that target dynamic networks, that are not always fully connected. We base our work on the Fractal component model, associated with its reference implementation Julia. The objectives of project Cubik are in particular to define a distribution method for a component in a dynamic network and to design some middleware support for the deployment and the distributed execution of the components.

Our work concerns mainly the following topics:

• Distributed execution of a composite software component
  Our approach consists in making the set of the component interfaces accessible on all of the hosts (or a subset of them) in the distributed platform, thus providing a ubiquitous component. On each of the hosts is also maintained information about the way subcomponents are distributed.
• Support for disconnections
  We assume that the machines that collectively host a composite component are not permanently connected. Hence it may occur that some of the interfaces of a component cannot be invoked sucessfully during a given period. Our approach consists in endowing the Fractal component model with states of interfaces (active or inactive). These states can be introspected through a control interface of the component. The runtime support makes it possible to reflect the state of network connections at the component level on the state of the interfaces.
• Deployment
  We focus on the last phases of the deployment, namely the instantiation and the activation of primitive and composite components on the hosts. This phase must take into account the heterogeneity of the hosts and of the links between them, and cope with possible disconnections. Our approach consists on one hand in enriching the architecture descriptor associated with the components, thanks the specification of constraints put on hosts resources and instances co-location; and on the other hand, in performing the deployment in a progressive way, by instantiating components as connectivity allows the constraints expressed in the deployment descriptor to be satisfied. The fact that interfaces can be made inactive permit the partial use of a component still being deployed.

Publications