嵌入式网络分布式负载任务并行计算系统设计

Design of embedded network distributed load task parallel computing system

传统的嵌入式网络分布式负载任务并行计算系统工作性能差。针对这一问题,设计一种新的并行计算系统,系统硬件由存储层、缓存层、执行层组成架构。通过STORE selection模块连接各层数据,Zookeeper模块监控系统的整体运行变化,data loader模块负责引入外部数据,利用分布式索引结构将得到的数据存储在系统中,设计失效节点检测机制增加系统的容错性,每个节点都设计对应的常驻进程,优化构建查询模块,提高用户查询效率。软件由准备阶段、执行阶段和完成阶段组成,每个阶段的任何一个环节出现问题,都会直接导致进程失败。设计对比实验检测系统效果,结果表明,相较于传统系统,文中给出的并行计算系统在计算过程中消耗的索引量更少,并行计算性能更强,所有的计算都可以在系统中进行,确保系统的负载均衡特性得以提高。

The performance of traditional embedded network distributed load task parallel computing system is poor.In view of this,a new parallel computing system is designed.The system hardware is composed of the storage layer,buffer layer and execution layer.The data of each layer is connected by STORE selection module.The Zookeeper module monitors the overall operation change of the system.The data loader module is responsible for introducing external data.The data obtained is stored in the system with distributed index structure.The failure node detection mechanism is designed to increase the fault tolerance of the system.The corresponding resident process is designed for each node to optimize the constructed query module and improve the query efficiency of users.Any problem in any link of each stage will directly lead to process failure.Problems in any step of each stage will directly lead to process failure.Contrastive experiments were designed to test the effect of the system.The results show that,in comparison with the traditional system,the proposed parallel computing system consumes fewer indexes and has more powerful parallel computing performance,so all calculations can be performed in the system,which ensures an improved system load balancing performance.