【摘要】 针对传统编码分布式计算方案中Shuffle阶段数据交换量过大导致的系统时延高以及可靠性低等问题，提出了一种基于三模冗余的编码分布式计算方案(Triple Modular Redundancy-Coded Distributed Computing,TMR-CDC)。首先创建了TMR-CDC模型，该模型将同一数据在不同节点进行冗余备份，在保证有足够中间值计算最终结果的同时减少数据传输量。其次在数据传输过程中融入校验机制检测计算结果的正确性，并设计了TMR-CDC算法实现该方案。最后通过实验表明，TMR-CDC方案将Shuffle阶段的运行时间占编码分布式计算总运行时间的比值从C-CDC方案的34.53%和S-CDC方案的33.27%降低到26.39%。并且TMR-CDC方案设置的容错机制显著提高了系统的可靠性。更多还原

【Abstract】 Aiming at the problems of high system delay and low reliability caused by excessive data exchange in the Shuffle phase in traditional Coded Distributed Computing schemes,a TMR-CDC scheme is proposed.Firstly,the TMR-CDC model is created,which backs up the same data redundantly at different nodes,and reduces the amount of data transmission while ensuring that there is enough intermediate value to calculate the final result.Secondly,the verification mechanism is integrated into the data transmission process to detect the correctness of the calculation results,and the TMR-CDC algorithm is designed to implement the scheme.Finally,experiments show that the TMR-CDC scheme reduces the ratio of the running time of the Shuffle phase to the total running time of the Coded Distributed Computing from 34.53% of the C-CDC scheme and 33.27% of the S-CDC scheme to 26.39%.Moreover,the fault-tolerant mechanism set in the TMR-CDC scheme significantly improves the reliability of the system.更多还原