欢迎您！东篱公司
退出
-
申报数据库
-
立项数据库
-
成果数据库
-
项目获奖数据库
中文摘要：
Join-Matrix是一种高性能的连接矩阵模型,方便部署于分布式环境下,支持任意连接谓词的数据流连接操作.由于采取随机分发元组作为路由策略,Join-Matrix可利用对元组内容的不敏感性来有效抵御数据倾斜.为了实现工作节点的负载均衡以及网络传输代价的最小化,基于连接矩阵模型设计一种高效的数据划分方案尤为重要.针对数据流连接处理,本文设计并实现了一种新颖的连接算子,可灵活地进行划分方案的自适应调整,以应对实时动态变化的数据分布.具体来说,我们根据数据流流量的采样信息和系统额定负载,通过一个轻量级的决策器制定出一个数据划分方案和相应的数据迁移计划,在保证输出结果完整性与正确性的情况下,实现迁移代价的最小化.本文在多种不同的数据集上进行了大量对比实验,结果证明,在资源利用率、系统吞吐率与时间延迟等方面,该连接算子较对比系统具有更高的性能体现.
英文摘要：
Join-Matrix is a high-performance model for stream join processing in a parallel shared~nothing environment, which supports arbitrary join operations and is resilient to data skew for taking random tuple distribution as its routing policy. To evenly distribute workload and minimize network communication cost, designing an efficient partitioning policy on the matrix is particularly essential. In this paper, we propose a novel stream join operator that continuously adjust its partitioning scheme to real-time data dynamics. Specifically, based on the sample statistics of streams and rated load of each physical machine, a lightweight scheme generator produces a partitioning scheme; then the corresponding solutions for state relocation are generated by a migration plan generator to minimize migration cost while ensuring result correctness.Our experiments on different kinds of data sets demonstrate that our operator outperforms the static-of-the-art strategies in resource utilization, throughput and system latency.
同期刊论文项目
同项目期刊论文
期刊信息
位置：