配电网监测数据微批处理的血统链标记容错法

针对分布式配电自动化系统存在数据量井喷、海量监测数据缺乏高效的分布式故障容错机制的问题,提出一种血统链标记容错新方法。利用弹性分布式数据集、微批计算的记录级容错和血统链标记序列融合处理的设计技巧,实现了分布式数据容错中血统链的追溯和条件标记的自动选择。以铁路配电网监测采集的数据为算例,搭建了4机集群的调度监控平台进行容错测试。以发生频次最高的单数据节点故障为例,测试结果表明:对于包含3×10~6条监测数据记录的弹性分布式数据集,血统链标记容错模型的集群CPU平均占用率波动小于1.5%,磁盘占用率下降4.2%;当迭代次数达到600、800次时,迭代运算耗时分别降低24.3%和42.9%;所提方法实现500 ms流处理延时的同时,对集群资源的使用情况也具有较好的优化效果,验证了该方法对分布式集群容错的有效性。

Lineage chain mark fault-tolerant method for micro-batching of distribution network monitoring data

Aiming at the existing problems of monitoring message blowout and lacking of efficient distributed fault-tolerant mechanism for huge amounts of monitoring data in the distributed distribution automation system, a novel lineage chain mark fault-tolerant method is proposed. The retrospection of lineage chain and the automatic selection of conditional marks in the distributed data fault-tolerant progress are realized by applying the design technique of fusing the resilient distributed data set, the record-level fault-tolerant of micro-batching computing and the lineage chain mark sequence. The monitoring data of railway distribution network are taken as the simulation example and the dispatch monitoring platform of a 4-generator cluster is built to carry out the fault-tolerant test. A single-node fault with the highest frequency is taken as an example, whose test results show that, for the resilient distributed data set including 3×106 monitoring data records, the average CPU utilization rate fluctuation is less than 1.5% and the disk utilization rate decreases by 4.2%;when the number of iteration increases to 600 and 800, the time-consuming decreases by 24.3% and 42.9% respectively; the proposed method not only realizes the stream processing delay of 500 ms, but also has a better optimization effect on the usage of cluster resources, verifying the effectiveness of the method for distributed cluster fault-tolerant.