FRESCOR

Scientific and technological OBJECTIVES

The main objective of the project is to develop the enabling technology and infrastructure required to effectively use the most advanced techniques developed for real-time applications with flexible scheduling requirements in embedded systems design methodologies and tools, providing the necessary elements to target reconfigurable processing modules and reconfigurable distributed architectures.

The approach to achieve this main objective is to integrate advanced flexible scheduling techniques directly into an embedded systems design methodology, covering all the levels involved in the implementation, from the OS primitives, through the middleware, up to the application level.

The most promising design methodologies for embedded systems, which provide the most reusability and the smallest time to market, are based on component-based techniques in which each component is independently developed, modeled, and tested. One of the challenges that these methodologies face when applied to embedded systems is the need to model the timing behavior of the component is such a way that it allows the application developer to independently analyze its timing properties, and to determine the timing properties of the overall system when integrating different components. But this timing behavior is heavily dependent on the execution platform and its particular scheduling policies, making it very difficult to separate the model from the surrounding environment. Therefore, a key issue is to create the infrastructure required to fully understand the overall timing properties of the system during the integration phase.

One very promising approach in this context is the use of the contracts specified in the Flexible Scheduling Framework (FSF) developed in the FIRST project (IST-2001 34140). This contract provides a model of what the application requirements are with respect to the flexible use of the processing resources in the system, and also of what are the resources that the system must guarantee if the component is installed into the system, and how the system can distribute any spare capacity that it has, to achieve the highest usage of the available resources. This contract-based approach allows the required level of abstraction to make the component model independent of the underlying implementation and hardware architecture. FSF has proven that it can be portably used across different scheduling strategies and platforms.

FSF represents just an initial step. To be able to integrate a scheduling framework like FSF into a component-based design methodology, several issues need to be addressed at different system levels, and constitute the specific objectives of this project.

1. At the application domain level, provide a quality of service manager that understands about the quality concepts of the application and translates them to the scheduling domain information that the underlying system can understand and implement, generating the contracts in an automatic or semiautomatic way. Of course application level quality varies from one application to another so this quality manager has necessarily a generic interface, and needs specific instantiation for a particular application. The manager provides a common reference architecture in which the core operations can be shared.

2. At the scheduling domain, develop a manager capable of managing the contracts of a distributed transaction, in distributed reconfigurable architectures. The current FSF infrastructure works on uniprocessor and distributed systems, but in the latter case the decisions are mostly local. A general truly distributed manager needs to be implemented to dynamically manage transactions on a global system basis.

In addition, we have identified several areas that are not supported in FSF and that would require attention in this project to make the resulting framework usable in a component-based development methodology that integrates all the system resources. These are specific objectives related with these areas:

3. Provide an integrated view of all the resources: in addition to the processor and network resources, there are many resources that should become part of the set with real-time scheduling requirements,to achieve full independence of the components used to build an application. The new framework will include all system resources in addition to the threads and networks: dynamically reconfigurable modules, interrupts with time protection, shared resources with time protection, memory protection, and energy/power-aware scheduling. Of course it is necessary to distinguish those resources that we can manage for scheduling purpose from those that we can just model, or that do not require specific scheduling.

4. Develop real-time network support: Different networks and protocols will be supported, including wireless networks, which are becoming so important in embedded systems today. In these networks we can build on the reservation mechanism developed in FSF, and add the ability to automatically distribute spare network capacity. In addition to using specific real-time networks and protocols, existing protocols with QoS capabilities should be addressed.

5. Develop enhancements that allow for more flexibility, better admission tests, and more scheduling theory that would allow better independent analysis of the component models. Using these results we propose building methods to perform contract calculations automatically, from the real-time requirements.

6. Extend the framework to new platforms, in particular multiprocessor and multi-core processor platforms, as well as mixed platforms including processors and special-purpose hardware reconfigurable co-processors built with FPGAs. This would allow following the current trends in processor and system HW architectures. The hardware platform models should be realistic, and non-CPU-centric, taking into account issues such as energy, bus management, coprocessors, and caches.

7. Use a tracing mechanism that would obtain the relevant execution time estimates and feed the obtained timing metrics into the design models. Other resource usages could also be traced, such as memory allocation needs.

8. Provide performance and quality evaluation via simulation of a system from the models of the components used, possibly enhanced with the timing metrics obtained. This is essential to evaluate the performance and achieved level of quality, which are parameters that are not obtained with the hard-real-time schedulability analysis tests.

9. Provide methods and tools to allow a design space exploration, using the results of performance and quality evaluation to make automatic or semi-automatic design decisions that allocate application components to resources and optimize the system’s behavior.

10. Address specific design tools used in control applications. In particular, the project will interface the developed framework with Matlab/Simulink-based design and development processes, as this opens the door to a large basis of potential users. The design should model the timing requirements and automatic code generation techniques would be used to generate the system.

It is necessary to get hands-on experience on the integration of this flexible scheduling technology into a particular component-based design methodology for reconfigurable embedded systems. Therefore, the specific objective in this area is:

11. Integrate an existing component-based framework with the underlying services required for resource scheduling providing different kinds of flexible real-time requirements to the component designer and the application developer.

The final specific objective will allow an overall evaluation of the framework:

12. Test the results of the project for three application domains, to evaluate the approaches, their effectiveness, efficiency, and applicability. These three domains are multimedia phone video streaming, software defined radio (SDR), and digital video surveillance, as major embedded systems application areas that are good examples of flexible timing requirements.