一种减少长尾延迟的分布式实时约束传播方法

发布时间：2018-03-28 03:29

本文选题：实时约束传播　切入点：长尾延迟　出处：《计算机研究与发展》2017年07期

【摘要】：提出了一种在数据中心环境下用于减少长尾延迟的分布式实时约束传播方法,该方法能够使当前节点感知请求的全局响应时间约束信息,并能够将请求的实时约束信息传播到整个处理路径;节点可以利用请求的实时约束信息进行请求调度或加速请求执行时间,以此来减少长尾延迟现象.同时,针对划分/聚合模式和串行/依赖模式2种数据中心应用,提出了阶段服务模型和并行单元模型,并基于这2种模型实现了分布式实时约束传播框架.最后,在分布式实时约束传播框架上实现了实时约束感知调度算法,通过实验进行了简单的验证,初步的实验结果显示了分布式实时约束传播方法能够在一定程度上减少长尾延迟.
[Abstract]:In this paper, a distributed real-time constraint propagation method for reducing long tail delay in data center environment is proposed. This method can make the current node aware of the global response time constraint information of the request. The real-time constraint information of the request can be propagated to the whole processing path, and the node can make use of the real-time constraint information of the request to schedule the request or accelerate the execution time of the request, thereby reducing the phenomenon of long tail delay. For two data center applications, partitioning / aggregation mode and serial / dependency mode, a phase service model and a parallel unit model are proposed. Based on these two models, a distributed real-time constraint propagation framework is implemented. A real-time constraint aware scheduling algorithm is implemented in the framework of distributed real-time constraint propagation. The experimental results show that the distributed real-time constraint propagation method can reduce the long tail delay to some extent.
【作者单位】：中国科学院计算技术研究所;中国科学院大学;
【基金】：国家自然科学基金国际合作项目(61420106013) 国家重点研发计划项目(2016YFB1000201)~~
【分类号】：TP308

【参考文献】