Section: New Results

Hardware Distributed Control for Dynamic Reconfigurable Systems

The progress in FPGA technology has allowed FPGA-based reconfigurable embedded systems to target increasingly sophisticated applications, which leads to a high design complexity of such systems especially at the adaptation control level. This complexity results into long design phases and delayed time-to-market. In this context, a centralized control model might be not adapted to the growing size and complexity of embedded systems. The use of a single controller for the whole system might result into a high complexity due to the number of parameters to take into account for runtime adaptation, which makes difficult its modification and test. Besides, the design of such a controller is system-dependent since it treats the system as a whole, which represents an obstacle for design reuse. In order to solve these problems, we propose a control design approach aiming to decrease design complexity and enhance design flexibility, reuse and productivity. This approach is based on a semi-distributed control model [34] . In order to achieve the objectives mentioned above, the proposed approach combines autonomy, modularity, formalism and high-level design. The semi-distributed control model divides the control problem between autonomous controllers handling each the self-adaptation of a reconfigurable component of the system, which allows to decrease their design complexity. Each controller handles three main tasks allocated to three different modules: i)monitoring of events that might trigger the adaptation of the controlled component, ii)decision-making about the required adaptations, and iii)adaptation (reconfiguration) realization. To ensure that reconfiguration decisions made by the controllers respect global system constraints such as security and quality of service constraints, these decisions are coordinated before launching the corresponding partial reconfigurations. The allocation of these tasks to separate modules facilitates their modification and reuse and thus the scalability of the control design. For the decision-making modeling, we use the mode-automata formalism. This formalism is suitable to model the control of the different modes of a reconfigurable system such as energy modes or image display modes. Thanks to its clear semantics, the use of such a formalism facilitates the high-level modeling of the controllers and their automatic generation. In order to facilitate code generation and enhance thus design productivity, our control approach makes use of Model-Driven-Engineering (MDE) [33] . Control systems composed of controllers and coordinators are modeled using the UML (Unified Modeling Language) profile MARTE ( Modeling and Analysis of Real-Time and Embedded systems). The automation of MDE, allowed to generate the code of these systems. The generated code was then used to validate the semi-distributed control and to determine its resource overhead compared to centralized control systems.