• Coordinator: P. Sens (LIP6/SU).
• Participants: K. Altisen (Verimag/UGA), A. Beynier (LIP6/SU), E. Caron (LIP/ENS Lyon), S. Devismes (Verimag/UGA), L. Lefevre (LIP/Inria), P. Sens (LIP6/SU), and hired Engineer 1.

Description:

The aim of this workpackage is to design the architecture of the SkyData environment. After building the SkyData architecture, we will build a prototype, and it will be used across the use cases created in Architecture and Protoype. Nevertheless, this prototype is neither a software, nor a middleware: we want to build a complete environment. A set of mechanisms are required to build this environment. We will consider five main modules.

Creation Module: We need to offer a solution to create Self-Managed Data according to WP1. In this step, we need to find how to extend the data with the required information to create the Self-Managed Data. Moreover the intelligence and skills of the data must be embedded in and will be managed through the runtime and service management modules. Runtime Module: The target of this module will be to provide mechanisms to execute the codes required to operate the Self-Managed Data. How to start (or launch) autonomous data? How to manage Self-Managed Data life cycle?

Service Management Module: In the SkyData environment, we want to provide large and dynamic capacities for the Self-Managed Data. We think that it will be very important to have a high granularity. Thus, we plan to study how to offer mechanisms for Self-Managed Data to deal with microservices. This module will take benefit from the results obtained in WP2 and WP4. Learning Module: This module will provide mechanisms for Self-Managed Data intercommunication. Beyond this communication protocol, we want to design a solution for Self-Managed Data to share their knowledge with each other in order to improve over time. This learning module should be done according to WP2 (Task 2.3). For example, multi-agent deep reinforcement learning might be one way to build an intelligent SkyData system.
Behavior Module: Based on the algorithms provided by WP2 and the inputs from WP4, we will evaluate in a real environment the behavior of Self-Managed Data. We want to generate some scenarios to observe the behavior of Self-Managed Data and validate the quality of the solution. We plan to investigate whether agent creation platforms could provide some foundations to our prototype. Our motivation is to reuse existing tools which could provide some state-of-the-art functionalities facilitating the creation and execution of complex reasoning applicable to autonomous data. However, autonomous data have to be easy to handle. For this purpose, we will pinpoint the level of functionalities which best meets the expectations. We preliminary considered JADE, 2 and JASON 3 . JADE is a middleware that implements a distributed multi-agent framework while JASON is an interpreter for AgentSpeak, one of the most common language for the Belief-Desire-Intention architecture. Both JADE and JASON are OpenSource project under the LGPL Licence.

We used JASON on preliminary work on the first version of a prototype and we aim to use the JADE infrastructure to make the prototype run on a large-scale network. Nevertheless, we also studied some other frameworks such as MARS 4 that uses agents to represent real entities.