• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文期刊>Computers and Artificial Intelligence >MANAGING PARALLELISM IN PARALLEL SYSTEMS
【24h】

MANAGING PARALLELISM IN PARALLEL SYSTEMS

机译：在并行系统中管理并行

获取原文
获取原文并翻译 | 示例
           
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

A fundamental problem of parallel computing is that applications often require large-size instances of an algorithm while parallel systems are generally hardwired architectures which cannot be easily reconfigured according to program size or structure. The unfolding of parallelism in an application can result in very large temporary storage requirements during periods in which results are being produced at a faster rate than instructions are being executed. This is an area of major concern when designing parallel systems. Strategies for managing parallelism have the basic aim of generating enough parallelism to utilize the machine fully, while at the same time keeping resource demands low enough to allow programs to execute without running out of storage resources. We first examine the parallelism management approaches adopted in a number of parallel computers such as the NEC μPD7281 Dataflow Chip, the FLAGSHIP computer, the NYU Ultracom-puter, the Zero Assignment Parallel Processor, the Rediflow Multiprocessor, the ETL SIGMA-1 Dataflow Machine and the MIT Tagged-Token Dataflow Machine. We then discuss the design of the throttling mechanism for managing parallelism in the Multi-Ring Manchester Dataflow Machine and analyzed the simulation results obtained. We conclude by refuting the criticism that parallel systems require excessive amount of memory resources for executing programs.
机译：并行计算的一个基本问题是，应用程序通常需要算法的大型实例，而并行系统通常是硬连线架构，无法根据程序大小或结构轻松进行重新配置。在应用程序中并行性的发展可能会导致非常大的临时存储需求，在此期间，生成结果的速度要快于执行指令的速度。在设计并行系统时，这是一个主要关注的领域。管理并行性的策略的基本目标是生成足够的并行性以充分利用计算机，同时将资源需求保持在足够低的水平，以允许程序在不耗尽存储资源的情况下执行。我们首先研究了许多并行计算机（例如NECμPD7281数据流芯片，FLAGSHIP计算机，NYU超计算机，零分配并行处理器，Rediflow多处理器，ETL SIGMA-1数据流计算机和MIT标记令牌数据流计算机。然后，我们讨论了多环曼彻斯特数据流机中用于管理并行性的节流机制的设计，并分析了获得的仿真结果。最后，我们驳斥了批评，即并行系统需要过多的内存资源来执行程序，因此得出结论。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号