4 Achieving Pbkp_redictable Performance in Multicore Embedded Real-Time Systems

Abstract:

4.1 Purpose The community at large has already recognized the risk involved in the trend to base the speedup of processors on an increasing number of cores instead of faster single cores. To harvest such a speedup enough executable elements (tasks, processes, threads, instructions depending on the level of parallelism desired) to be executed in parallel in each of the cores need to be found. New allocation, scheduling and synchronization techniques are needed in order to utilize the capacity of the cores. There are two aspects:(1) ensuring that the system maintains its pbkp_redictable timing behavior when executing on a multicore, and (2) maximizing the parallelism in order to minimize idleness of some of the cores. It is paramount that we get a pbkp_redictable timing of such execution to be able to guarantee deadlines. For any power-constrained or energy-conserving embedded systems it is also essential that no core is idly consuming energy (one can adjust and voltage scale a multicore chip, but not necessarily turn off a single core). On top of this, the consequence of wasting processor performance in embedded systems thus can have a higher penalty provided that cost is an important driving force in embedded systems as is the case in, for example, automotive systems, where economy of scale magnifies the cost of hardware underutilization. As a result, missing the performance improvements of multicore systems in embedded systems can have huge economic consequences. Current multicore chips contain typically four to eight homogeneous cores. Within three to five years, multicore chips will exceed 32 cores, and also be heterogeneous, that …

Año de publicación:

2009

Keywords:

Fuente:

google