...
【24h】

Don't Drop, Detour!

机译:别摔,绕路!

获取原文
获取原文并翻译 | 示例

摘要

Modern data center networks (DCNs) are built with shallow-buffered switches. These switches are cheaper than their deep-buffered counterparts, and they also reduce the maximum queuing delays packets may suffer. However, shallow buffers can lead to high packet loss under bursty traffic conditions (often caused by the "incast" traffic pattern [7]). DCTCP [3] attempts to solve this problem by using ECN markings to throttle the flows early, to avoid buffer overflow. However, DCTCP cannot prevent packet loss if the traffic bursts are severe, and short-lived. One extreme example is a large number of senders each sending one or two packets to a single receiver. No feedback-based congestion control protocol can prevent buffer overflow and packet losses under such extreme conditions. One reason for this is an assumption ingrained in today's DCN design.
机译:现代数据中心网络(DCN)内置有浅缓冲交换机。这些交换机比它们的深度缓冲交换机便宜,而且还减少了数据包可能遭受的最大排队延迟。然而,在突发流量情况下(通常由“ incast”流量模式[7]引起),浅缓冲区会导致高丢包率。 DCTCP [3]试图通过使用ECN标记来尽早限制流量来解决此问题,以避免缓冲区溢出。但是,如果流量突发严重且持续时间很短,则DCTCP无法防止数据包丢失。一个极端的例子是大量发送器,每个发送器向一个接收器发送一个或两个数据包。在这种极端条件下,没有基于反馈的拥塞控制协议可以防止缓冲区溢出和包丢失。原因之一是当今DCN设计中根深蒂固的假设。

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