Reliability Improvement For Many-core System Based On Processing Core Apportion Algorithm

Space vehicles work in extreme environment of outer space where thepossibility of all kinds of function failure will increase dramatically. Withthe electronic devices integration level increasing, the feature size ofelectronic units decreases. This trend on one hand makes electronic unitsmore sensitive to space radiation which will lower the reliability ofelectronic devices in space vehicles further more. On the other hand, wecan integrate more than one processing core in a single processing IC. Theincrease of processing cores in a single processing IC not only improvesthe performance in a relatively low energy cost, but also provides thepossibility for the control system to apply reliability improvement methodsin module level. Thus the application for many-core system in spacevehicle is definite.Our research focuses on the way to apportion the limited resources inmany-core system in order to achieve the highest reliability.The apportionment algorithms we currently use have severallimitations. Firstly, the current used apportionment algorithms are used indeveloping phase of an electronic device, especially in design stage andcannot be applied in working phase after delivering to users. Secondly, thecurrent used apportionment algorithms are neither fast nor precise. Thispaper will propose a new apportionment algorithm which will be improvedin the two mentioned aspects, which will be named as―influence-firstapportionment algorithm‖.For a many-core system, it can apportion its processing coresdynamically and in real-time due to the use of Partial Dynamic Reconfiguration (PDR) technology. Instead of the current traversalalgorithms, this paper proposes a sensitive-first apportionment algorithm.Method to decide the degree for a subsystem influencing total system isthus proposed. What’s more, we also propose a new evaluation method toevaluate the availability of the system. When considering apportioning fast,dynamic and in real-time, the evaluation method must be fast enough whilekeep precise. Basically, the proposed evaluation method is an improvementof the currently used Markov Model Method. Then mathematical analysisis applied to prove the accuracy of the proposed method and simulationexperiment is applied to verify the time-complexity of the proposedapportionment algorithm.