面向大数据的高性能网络带宽调度与传输控制关键技术研究

项目名称： 面向大数据的高性能网络带宽调度与传输控制关键技术研究

项目编号： No.61472320

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 自动化技术、计算机技术

项目作者： 吴奇石

作者单位： 西北大学

项目金额： 84万元

中文摘要： 网络环境下的大规模数据处理应用需要海量数据传输，而传统的IP网络难以胜任。具有高带宽及带宽预留特点的高性能网络（HPNs）是一种公认的有效方案。然而，HPNs通常由不同地域用户分时共享，其数据传输性能严重依赖于主机传输协议。为此，迫切需要研究高效带宽调度和自适应传输控制方法。本项目旨在研究透明、统一的主机-网络接口，基于事件驱动控制流的用户、系统和网络实时运行参数动态调整与响应方法；研究传输控制感知的即时带宽和周期性带宽调度算法及其高效启发式算法，解决不同网络条件和传输需求下动态传输带宽分配，即何时传的问题；研究带宽调度感知、吞吐量最大化的批量数据传输协议和性能自适应的网络传输控制方法，解决怎样传的问题。项目探索将HPNs中带宽调度和终端系统中传输控制紧密结合、控制平面和终端主机之间互相协同工作的高效数据传输方法与协议，研究结果对高性能网络和海量数据科学应用的性能优化有着重要意义。

中文关键词： 大数据；高性能网络；带宽调度；传输控制；性能模型

英文摘要： Many large-scale applications in different domains are generating colossal amounts of data, which must be transferred over long distances for remote operations. As the conventional best-effort IP networks are not adequate to meet such unprecedented data transfer challenge, high-performance networks (HPNs) featuring high bandwidth and advance reservation have emerged to be a promising solution to support these data- and network-intensive applications. The bandwidths in HPNs are typically time-shared among geographically distributed users, and the user-level throughput performance largely depends on the specific transport method being used, which necessitates the research on a holistic data transfer solution that combines both bandwidth scheduling (BS) in the network and transport control (TC) on the host. This project will develop a transparent and unified host-network interface that utilizes an event-driven control flow scheme for runtime adaptation in response to user, system, and network dynamics. We will design TC-aware BS algorithms to determine when to transfer under disparate network conditions and user requirements, and BS-aware performance-adaptive TC methods to determine how to transfer with different goals of throughput maximization and stabilization, both of which are seamlessly integrated in a unified framework. The completion of this project will result in a set of novel cooperative scheduling algorithms and transport protocols that can coordinate with each other between control planes and end hosts to optimize the overall network and application performance.

英文关键词： Big data;high-performance networks;bandwidth scheduling;transport control;performance modeling