Data Center Specific Networking Components

数据中心特定网络组件

• 批准号: 519969-2017
• 负责人: Langlois, Pierre
• 金额: $ 1.82万
• 依托单位: École Polytechnique de Montréal
• 依托单位国家: 加拿大
• 项目类别: Engage Grants Program
• 财政年份: 2017
• 资助国家: 加拿大
• 起止时间: 2017-01-01 至 2018-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=639607
• 关键词: Data; Center; Specific; Networking; Components

The goal of this project is to exploit the algorithms developed by the team in which Prof. Langlois is a memberto fit in Kaloom's environment, to assess their adequacy and to develop new versions of these algorithms inorder to achieve the performance required by our industrial partner. The focus will be on the specific needs ofdata centers and the research will revolve around improving the algorithms to better fit those needs. Thescientific issues and research problems relate to typical problems in computer architecture and to the specificnature of network traffic processing. One major scientific issue is to support very large routing tables that arelarger than integrated circuits can accommodate. The size of those tables prevents using hardware-intensivesolutions like TCAMs because the hardware cost would be prohibitive. Also, software-based solutions aremuch too slow for current interconnects (up to 100 Gb/s per link). Moreover, current state-of-the-art externalmemories (DRAM and SRAM) are too slow to meet the needs. Thus, new algorithms must be devised to allowmost operations to be performed using on-chip data and to resort to external memory only very infrequently.

该项目的目标是利用 Langlois 教授所在团队开发的算法来适应 Kaloom 的环境，评估其充分性并开发这些算法的新版本，以实现我们的工业合作伙伴所需的性能。重点将放在数据中心的特定需求上，研究将围绕改进算法以更好地满足这些需求。科学问题和研究问题涉及计算机体系结构中的典型问题以及网络流量处理的具体性质。一个主要的科学问题是支持超出集成电路所能容纳的非常大的路由表。这些表的大小阻碍了使用 TCAM 等硬件密集型解决方案，因为硬件成本过高。此外，基于软件的解决方案对于当前的互连来说太慢了（每个链路高达 100 Gb/s）。此外，当前最先进的外部存储器（DRAM 和 SRAM）速度太慢，无法满足需求。因此，必须设计新的算法，以允许使用片上数据执行大多数操作，并且仅很少使用外部存储器。