Towards Distributed Robustness in Embedded Systems

Recently, we witness a significant amount of effort being put into a research of robustness of complex computer systems. Computer scientists noticed that, despite decades of investigations into dependability, computer systems still lack the degree of resilience which can be seen in biological systems. In the biological context, one of the main aspects that stand behind a system’s robustness is degeneracy [1] also known as distributed robustness [2]. We say that a system has a high degree of degeneracy if a high number of the system’s parts functionally overlap. A system designer can harness degeneracy and provide fault-tolerance through procedural (in contrast to structural) redundancy. A system element is structurally redundant if another element in the system provides identical functionality. However, an element is procedurally redundant if various system parts can interact in such a way that it is possible to reproduce the element’s functionality. For illustration, consider a hypothetical calculator which was designed for degeneracy. Such a calculator does not have a central CPU, but its functionality is rather distributed among number of function blocks which provide basic arithmetic operations (plus, minus, multiplication, and division). If one of the function blocks is affected by a permanent error, there is no other functionally identical element which could take over the lost function. Yet, the blocks functionally overlap. If, for example, the multiplication operator failed, the calculator could transform any multiplication into a sequence of additions and use the plus operator. If the plus operator failed, the minus operator can be used in a similar fashion. Dependable computer system of today are still clumsily “inorganic”, with a nonexistent or very low degree of adaptability. Procedural redundancy due to degeneracy is a promising way to achieve highly robust and adaptable systems. But designing for degeneracy, and that is where we want to make our point, requires a paradigm shift in the way computing is done. We argue for a novel model of computation which enables a design for degeneracy.