在多线程并行程序中,如只由主线程负责容错处理,主线程本身则成为潜在的单点故障点。针对该问题,设计一种对等多线程动态容错模型。该模型包含多轮容错过程,每轮容错过程采用容错控制线程随机决定容错模型动态生成的方法。在EasyPDP系统中进行实际应用测试,结果证明,该容错模型能够避免主线程单点故障,同时保证系统加速比与负载平衡性能。
In a multithread parallel program, the master thread is an underlying single point failure under the consumption that the fault-tolerance is handled by the mast thread only. To solve this problem, a Peer-to-Peer(P2P) multithread dynamic fault-tolerance model is designed and implemented. It contains several rounds of fault-tolerance process, fault-tolerance control thread is decided randomly and the fault-tolerance model is generated dynamically in every round. Test work is conducted in the system of EasyPDP, and results shows that this model is able to avoid the single point failure of the master thread, as well as maintain the system speed-up ratio and load balance performance.