Contents

  • Abstract
  • Dynamic Energy Management and Energy Centric Quantitative Model Checking
  • Objectives
  • Workplan
  • Abstract

    SENSATION aims at increasing the scale of systems that are self-supporting by balancing energy harvesting and consumption up to the level of complete products. In order to build such Energy Centric Systems, embedded system designers face the quest for optimal performance within acceptable reliability and tight energy bounds. Programming systems that reconfigure themselves in view of changing tasks, resources, errors and available energy is a demanding challenge.

    The lack of effective design-time support for taking on this challenge obstructs the creativity and productivity of design teams. This is an impediment to European companies developing embedded components, devices, and platforms, and is a major obstacle to developing self-supporting systems.

    SENSATION will free the system design process by devising energy-centric modeling and optimization tools for the design of resource-optimal reliable systems. This depends on orchestrated, non-incremental progress in several research domains. The project combines Europe's leading scientists in model-based quantitative evaluation and optimization, and in low-power reconfigurable systems.

    SENSATION provides automated analysis and synthesis tools for energy-centric systems. For the first time, tools for optimizing performance and reliability will be integrated with energy analysis. Based on efficient model-checking algorithms and massive design space exploration, this leads to a many-fold increase in system design productivity.

    Two industrial partners, GomSpace and Recore Systems provide challenging case studies and serve as industrial testbeds. The yardstick for the impact of SENSATION is a reduction in energy consumption by 50% and a reduction in time-to-market of at least 10%. The Center for Embedded Software Systems (CISS) at Aalborg University will actively contribute to the development of the technology and its effective dissemination and industrial adoption.

    Dynamic Energy Management and Energy Centric Quantitative Model Checking

    Dynamic Energy Management, a novel system concept. The first big challenge is that the system's workload, energy yield, buffer and demand, its performance constraints and the occurrence of faults, change over time and cannot be predicted at runtime, let alone at design time.

    The principle of Energy Centric Systems is that, based on runtime input, e.g. required QoS, remaining energy, and based on its internal energy model, the system reconfigures itself to optimize energy usage. This means that scheduling decisions (assigning time slots to tasks) and mapping decisions (assigning tasks to heterogeneous cores) will be resolved using the energy model. But also, several energy saving measures must be applied at runtime. Typical measures are: selectively reducing a component's clock speed or lowering the voltage (up to deep-sleep / temporary switch-off), clever usage of batteries, and replacing algorithms by more efficient approximations.

    An even harder challenge is that measures to increase either energy efficiency, performance, or reliability will often have contra-productive effects on the other criteria. This asks for an integral approach, taking quantitative constraints from different domains, and making trade-offs by optimizing various constraints simultaneously, while respecting limiting constraints on the other criteria.

    Energy Centric Quantitative Model Checking, a novel design technique. One of the main concerns inSENSATION is how to design systems with complex dynamic energy management. Those systems must respond to unknown and unpredictable inputs and events, in an energy optimal manner, without compromising constraints on performance and reliability unnecessarily.

    In SENSATION the most recent techniques from Quantitative Modeling and Analysis will be advanced further and applied, for the first time, to the field of Energy Centric Systems.

    So far, quantitative modeling has been applied to complex systems with (conflicting) requirements on performance and reliability only. Those models are based on automata, which describe the allowed transitions between system states. These automata are extended with quantitative information, such as clocks (to model hard and soft real-time constraints), stochastic rates (to model arrival and failure rates, and express reliability constraints) and costs (to model limited computational resource budgets).

    In SENSATION energy will be incorporated in quantitative modeling frameworks and analysis algorithms to obtain Energy Aware Automata, where energy (unlike costs) can increase and decrease. These automata can be analyzed by means of model checking algorithms. These techniques can compute the average and maximum duration, probability, or cost of all system traces of a particular shape. By parameter sweeps, tradeoffs can be studied and used as input to the design space exploration.

    Objectives

    The overall objective of SENSATION is to enable the design of energy-wise self-supporting embedded systems up to the level of complete products. We envisage broadcast radio receivers that can function in the desert without any supplementary power source; security cameras that do not depend on external power cables, and deep space missions or satellites that keep working with minimal energy consumption.

    SENSATION targets the above objective with a particular aim at having an impact on the foundations of computing technologies with negligible energy consumption, striving towards enlarging the scope for energy-wise self-sustainable systems. In this context the main challenges and objectives for the SENSATION project are:

    1. To develop adequate dataflow and automata based modeling formalisms to describe a wide range of energy-related systems, and tailored towards power-aware optimization.
    2. To advance quantitative model-checking techniques and tools to allow for scalable model-based quantitative analysis of energy-aware models.
    3. To provide algorithmic and tool support for automatic synthesis of energy-optimal adaptive and dynamic energy management strategies.
    4. To provide a design exploration method allowing to analyse the effect of design choices in terms of a trade-off between energy, performance and reliability.
    5. To experimentally demonstrate the radically increased scale of systems being energy-wise self-supporting ranging based on cases arising from space missions, streaming applications and software-defined radios.

    Workplan

    The primary objective of SENSATION is to facilitate the design, analysis, and development of Energy Centric Systems through Dynamic Energy Management and Energy-centric Quantitative Model Checking. Achieving this goal depends on orchestrated, non-incremental progress in several research domains. The workplan to be implemented is split into six workpackages. The diagram below illustrates the interactions between workpackages,

    Workpackages and their interaction in SENSATION

    The three core technical work packages each correspond to a particular research domain and are well-connected with each other.
    1. Power Aware Modelling. The key to designing and developing effecient Energy Centric Systems, is sound, automated modelling of the system and its parts. WP1 will provide a solid foundation for the analytic work in SENSATION by defining a modelling framework for energy-aware system abstractions that enable a sound method for applying novel model checking and static analysis techniques. The work in this domain is led by Holger Hermanns (SAU).
    2. Scalable Evaluation and Optimization Techniques. WP2 will enable efficient scalable analysis and optimization of Energy Centric Systems by delivering novel quantitative energy centric model checking techniques that operate on sound abstractions of the components. The specification of the components and the system will rely on the framework provided by WP1. This work package will be led by Joost-Pieter Katoen (RWTH).
    3. Energy Aware Dynamic Resource Allocation. Realizing energy-efficient systems on dedicated execution platforms necessitates complete behavioral models of both the platform and the system. In order to achieve this, WP3 will combine the energy-aware modelling formalisms developed in WP1 with the analysis techniques developed in WP2 and apply these to the modelling and analysis of Energy Centric Systems to realize energy optimal systems. This work package will be led by Axel Legay (INRIA).

    The work in the three WPs above will be driven by the needs of our industrial cases, which are bundled in WP4, Demonstration and Validation. Novel techniques from WP1-3 will be demonstrated and evaluated on these case studies. This work package will be led by UT researchers Jaco van de Pol and Gerard Smit.

    The validation cases will also serve as excellent show-cases for subsequent demonstration and exploitation of the SENSATION technology. These activities are bundled in a dedicated work package, WP5, Dissemination, Exploitation, and Collaboration, led by Boudewijn Haverkort (ET). The dissemination activities thereby ensure maximum exposure to the end user community.

    The overall project will be coordinated by Kim G. Larsen, head of the Danish center for embedded software (CISS), based at Aalborg University (AAU).