结合指令预取和缓存划分的多核缓存WCEC优化

Instruction Prefetching and Cache Partitioning for Multicore Cache WCEC Optimization

对于能量供应有限制的硬实时多核系统,最差情况下的能量消耗WCEC(Worst-Case Energy Consumption)是一个非常关键的问题。随着芯片工艺的发展,顺序指令预取技术可以减少缓存WCEC。为了提高指令预取的最差情况下的节能效率,提出结合指令预取和共享缓存划分的硬实时多核系统缓存WCEC优化方法。该方法通过线性规划方程ILP(Integer-Linear Programing)为每个核分配L2缓存划分因子和调整每个硬实时子任务的指令预取度,在保证硬实时系统满足时间截止期的情况下,最小化其缓存WCEC。对DEBIE系统进行实例分析,实验结果表明优化方法是有效的,在保证系统满足时间截止期的情况下,优化后的缓存WCEC比没有指令预取优化的缓存WCET平均减少了22.5%。

For the hard real-time multicore system that its energy supply is limited,the cache WCEC(Worst-Case Energy Consumption)is a very critical problem.With the development of chip technology,sequential instructions prefetching can reduce the cache WCEC.In order to improve the worst case energy saving efficiency of instruction prefetching,the cache WCEC optimization method combined with the instruction prefetching and shared cache partitioning is proposed.The method assigns L2 cache partitioning factor for each core and adjusts instruction prefetching degree for each subtask by ILP(Integer-Linear Programing),to minimize the cache WCEC of hard real-time multicore system when guarantees the system meets the time deadline.The DEBIE system is analyzed in the experiments,and the results show that optimization method is effective,the optimized cache WCEC is decreased by 22.5%on an average than that of without instruction prefetching optimization when guarantees DEBIE system meet its time deadline.