首页> 外国专利> PCI Method for implementing barrier based on PCI Express in interconnection network

PCI Method for implementing barrier based on PCI Express in interconnection network

机译:在互连网络中基于PCI Express实现屏障的PCI方法

摘要

A method for implementing a barrier in a PCI Express-based connection network according to an embodiment of the present invention includes: transmitting a first count value, which is incremented when a DMA operation is performed in a PE (PCI Express) node at a user level, , The kernel level driver compares the second count value, which is a value obtained by incrementing the counter through the interrupt handler each time an interruption is generated after the completion of the DMA data transfer, with the first count value, And if the two values are the same, confirming that all DMA data transfers have been completed. If it is determined that all DMA data transfers have been completed, the kernel level driver instructs other PE nodes to send a door- Wherein the kernel level driver counts each time a barrier is executed, Calculating a first barrier count value by adding the number of other PE nodes except for itself to the barrier count value; and the kernel level driver increments the number of interrupts through an interrupt handler whenever another PE node generates an interrupt Comparing the first interrupt count value with the first barrier count value and waiting until the two values become equal to each other, and if the two values are equal, confirming that the barrier performance of all the PE nodes is completed. According to the present invention, it is possible to achieve extremely high-performance communication such as Infiniband and to reduce the cost.
机译:根据本发明实施例的用于在基于PCI Express的连接网络中实现屏障的方法包括:发送第一计数值,当在用户处的PE(PCI Express)节点中执行DMA操作时,该第一计数值递增内核级驱动程序将第二计数值与第一计数值进行比较,该第二计数值是在DMA数据传输完成后每次产生中断时通过中断处理程序使计数器递增而获得的值。两个值相同,确认所有DMA数据传输已完成。如果确定所有DMA数据传输均已完成,则内核级驱动程序指示其他PE节点发送一道门。其中,内核级驱动程序每次执行屏障时都会进行计数,并通过将第一个屏障计数值加除自身以外的其他PE节点的屏障计数值;每当另一个PE节点生成中断时,内核级驱动程序就会通过中断处理程序增加中断数量,将第一个中断计数值与第一个障碍计数值进行比较,并等待直到两个值彼此相等,以及两个值是否相等相等,确认所有PE节点的屏障性能均已完成。根据本发明,可以实现诸如Infiniband的极高性能的通信并降低成本。

著录项

  • 公开/公告号KR101948163B1

    专利类型

  • 公开/公告日2019-02-14

    原文格式PDF

  • 申请/专利权人 충북대학교 산학협력단;

    申请/专利号KR20170003333

  • 发明设计人 최민;

    申请日2017-01-10

  • 分类号G06F13/42;G06F9/50;

  • 国家 KR

  • 入库时间 2022-08-21 11:49:08

相似文献

  • 专利
  • 外文文献
  • 中文文献
获取专利

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