The Real-time Research On The Multi-processor Embedded System

Multi-core Processor currently popular in the server and desktop applications, will soon be widely used in high-end embedded field. The previous real-time embedded systems research focused on switching and task scheduling, real-time communications and fault-tolerance, not relate to real-time task itself. And multi-processor more concerned about the optimization of throughput in the system, rather than a real-time optimization.Based on the analysis of multi-processor characteristics and related work at home and abroad, this thesis researched the real-time problem by rebuilding the real-time tasks. Now multi-processors can share the 2-level cache and this property is utilized to further improve the performance of the real-time tasks.This thesis presents the real-time improvement scheme on multi-processor embedded system, which is composed by task parallelism and operating system support. Task parallelism produces multi-thread by means of functional decomposition and data composition. By comparing the performances before and after parallelism, this thesis concludes the reasonable decomposition to improve the real-time property of the task. Improvement scheme also includes 2-level cache based thread allocation algorithm, which supports the real-time task at run time.Experiments have proved that the real-time tasks are highly improved by the thesis's design. The testing environment is Simics simulator. The kernel of a media player is divided into three threads by functional decomposition and gets 102.6% speedup. Quick sort and merge sort are divided into two threads by data decomposition and get 87% and 92% speedup respectively. In this thesis's test case, 2-level cache based thread allocation can enhance the performance of 6.25%.Through the scheme in this thesis, the real-time task's performance is highly improved, thereby reducing the responding period.