| It has long been a tenet of the fault-tolerance research community that fault-tolerance is possible by exploiting resource redundancy in order to achieve the mission result. Hardware designers, for instance, rely upon various forms of redundant hardware to ensure the availability of at least one system that can produce correct responses. Strangely, one source of redundancy has not been comprehensively examined. Such redundancy is found in the optimizations used by system designers to enhance the core system mission. It is possible to design reliable systems that, upon detecting a fault, shed such optimizations. Most users are willing to forgo an optimization (such as a few percentage points of fuel efficiency) rather than face full system failure. Such a gentle system response to faults is the hallmark of a gracefully degrading system. Fault-tolerance based upon graceful degradation provides additional techniques to build dependable systems without the heavy costs of hardware replication.; This work identifies a heretofore-unknown problem in the system synthesis research space and explores solution methods. The problem, called system-wide customization, is to maximize the utility of a system with pre-specified hardware by selecting and allocating software components from an extensive, flexible library.; The importance and applicability of system-wide customization is apparent when focusing on the distributed embedded system domain. Within this domain, fixed hardware resources—distributed microcontrollers and the networks connecting them—limit the software components that can be executed. Each choice of possible hardware components can be viewed as a single vertex of a dense lattice that represents a fine-grained product family architecture (PFA). For each combination of hardware components, there are many different software configurations available. Using the PFA lattice concept, the system-wide customization problem may be expressed as the process of choosing the software configuration for a particular vertex (e.g., the one representing available hardware) that maximizes the utility of the system. |
