期刊名称：International Journal of Computer Science and Network
印刷版ISSN：2277-5420
出版年度：2012
卷号：1
期号：5
出版社：IJCSN publisher
摘要：As embedded applications are processing increasingly larger datasets,keeping their memory space consumptions under control isbecoming a verypressing issue. Observing this, several priorefforts have considered memoryspace reduction techniques (inboth hardware and software) based on data compression andlifetime-based memory recycling. In this work,we propose andevaluate an alternate approach to memory space saving in multicoreembedded architectures such as chip multiprocessors. Theuniquecharacteristic of our approach is that it recomputes theresults of selecttasks in a given task graph (which represents theapplication), instead of storing these results in memoryandaccessing them from there as needed. Our approach can workunder a givenperformance degradation bound and reducesmemory space requirements underthis bound. Our experimentalresultsare very encouraging and show that the proposedapproachreduces memory space requirements of our task graphsby as much as 19.5%,the average savings being around 11.3%.
关键词：Memory Space Reduction; Task Recomputation;Multi-core Architecture; Embedded Systems
