A Coflow Classifier to Support the Flow Scheduling in Optical Burst Ring Network
首发时间:2017-12-06
Abstract:With the rapid development of big data services and distributed computing systems, their network communication performance has been approaching the bottleneck. Unlike the characteristic of end-to-end services in the traditional network, their communication features are characterized by multipoint to multipoint and strong performance association between flows. To solve the communication problem of distributed computing system network, we need to identify and classify these collaborative flows which are named as Coflow, a network abstract concept for collection of flows that share a common performance goal. Meanwhile, relying on the structural advantages of the optical burst ring network, information interacting with low control overhead, and the features of burst switching with fine-grained bandwidth adjustment, we can achieve flow scheduling eventually. Thus the basic optical burst ring network can achieve effective computing system interconnection. This paper aims to extend the optical burst ring network node with a Coflow classifier, so that it can support Coflow\'s classification and scheduling. And through building the FPGA-based network switching platform, our experiments have successfully verified the effectiveness of the extension design.
keywords: network communication system optical switching Coflow flow classification flow scheduling
点击查看论文中文信息
一种支持光突发环网中流调度的Coflow分类器设计与实现
摘要:随着大数据业务和分布式计算系统的快速发展,其网络通信性能已日益临近瓶颈。与传统网络中业务端到端特性不同,这些通信的特点体现为多点到多点且流之间具有极强的性能关联。解决分布式计算系统网络通信问题,就需要能够分类识别这些彼此协作的流。为了描述这些具有共同性能目标的协作流,它们被称作为Coflow,一个网络抽象概念。与此同时,依赖光突发环网在结构上的优势,通过低控制开销的信息交互,以及利用突发交换细粒度带宽调整的特点,最终实现协作流的调度,使基本的光突发环网能够实现有效的计算系统互联。本文的目标在于以Coflow分类器来扩展光突发环网节点,使其具备协作流分类及调度功能,并搭建基于FPGA的网络交换平台,通过实验成功验证了这种扩展改进设计的有效性。
基金:
引用
No.****
同行评议
勘误表
一种支持光突发环网中流调度的Coflow分类器设计与实现
评论
全部评论0/1000