众核平台的并行编程模型及其运行时支持技术的研究

项目名称： 众核平台的并行编程模型及其运行时支持技术的研究

项目编号： No.61272099

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 过敏意

作者单位： 上海交通大学

项目金额： 81万元

中文摘要： 摩尔定律在众核领域的延伸导致处理器核心的数目越来越多,晶体管的尺寸越来越小，从而不仅使众核架构下的软件生产遭受严重危机，而且也使众核系统的管理面临巨大挑战。软件危机来自于众核编程的困难，而管理挑战源于当前众核系统的能耗和可靠性已经上升到和性能同等重要，甚至更为重要的问题。传统的和最近的并行编程模型要么并非为众核处理器设计而难以解决编程的困难，要么未对能耗和可靠性进行充分考虑。而这些问题许多可以在运行时得到解决。为此，本项目将围绕高可用、高性能和低能耗三个关键问题，开展以下几个方面的研究：研究面向众核典型应用中的并行性与数据分布的内在机理，设计并实现面向众核的编程模型；研究相应的运行时支持，综合考虑性能、能耗和可靠性因素；研究众核平台中体系结构扩展对众核编程模型的支持。项目成果有望对面向众核处理器结构的并行编程模型及其运行时支持技术提供一个较为高效的解决方案。

中文关键词： 众核；调度；编程模型；高性能；高能效

英文摘要： To fulfil the increasing demand for high performance with low energy consumption (Flops per Watt), processors architecture is evolving towards many-cores. In the future, processors will include tens if not hundreds of cores on a single chip, connected through a network implemented on the chip. Programming models will have to be rethought to cope with such a high level of parallelism. Therefore, programming and producing large scale software become more difficult. Futhermore, the resource management and run-time support for the programming models on many-core chips meet the challenges, because the energy consumption and reliability are same important with performance for multi/many-core systems. This Project will research on the novel programming models and their run-time support techniques for multi/many-core platforms. Mainly, we hope to focus on models for hierarchical parallelism, heterogeneous self-adaptive, and automatical data distribution. At the run-time, some techniques such as dynamic detection and monitor, task stealing scheduling, power consumption management, and code transformation will be investigated and studied.

英文关键词： Many core；Scheduling；Programming Model；High Performance；Energy Efficiency