DRAM, SRAM

DRAM的访问时间为数十纳秒. [1]SRAM的访问延迟通常为20~40纳秒, 而DRAM的访问延迟通常为60~100纳秒.[2]个人电脑中DRAM的访问延迟为50~150纳秒, 而SRAM的访问延迟可低达10纳秒. 相比之下, 个人电脑的硬盘的访问延迟为9~15毫秒. [3]

数据中心网络

一个大型的数据中心网络一般有超过100,000台服务器, 每台服务器有10~40Gbps的外接带宽, 总带宽很容易超过100 Tbps. [4]数据中心网络中, 当使用TCP的时候, 端到端的传输延迟一般是2毫秒, 但是如果使用RDMA, 则端到端传输延迟可减低到20微秒. [6]根据阿里巴巴的记录, 86%的网络性能异常都是由丢包引起的. 此外, 拥塞引起的丢包并不是网络性能异常的原因. 事实上, 路由黑洞, ACL 策略, TTL为0, 数据包长度超过MTU等产生的丢包导致了60%以上的网络性能异常, 而拥塞产生的丢包只导致了10%的网络性能异常. [6]Windows Azure的网络包含超过7000台网络设备和22万条链路. Windows Azure通过这些网络设备和链路来连接它们的分布在全世界的服务器. [7]由于网络设备和链路发生故障而导致少量的丢包 (比如5%以下的丢包率) 是非常常见的, 比如设备维护和设备功能失常导致的丢包. 相比之下, 光纤损毁和设备彻底停止运行等事故是比较少见的, 比如在我们观察到的73件网络事故中, 仅有5件网络事故属于这种类型.

互联网

互联网中一个数据包在传播过程中, 所经过的跳数的中位数是12, 只有少量的情况下其跳数会超过30. [5]

其它

RDMA的MTU为1000字节. [5]Ethernet的MTU为1500字节. [5] $\text{参考文献}$ [1] Y. Lu, A. Montanari, B. Prabhakar, S. Dharmapurikar, and A. Kabbani, “Counter Braids: A Novel Counter Architecture for Per-flow Measurement,” in Proceedings of the 2008 ACM SIGMETRICS International Conference on Measurement and Modeling of Computer Systems, New York, NY, USA, 2008, pp. 121–132. [2] https://www.enterprisestorageforum.com/storage-hardware/sram-vs-dram.html [3] https://www.webopedia.com/TERM/A/access_time.html [4] Y. Zhou et al., “Flow Event Telemetry on Programmable Data Plane,” in Proceedings of the Annual conference of the ACM Special Interest Group on Data Communication on the applications, technologies, architectures, and protocols for computer communication, New York, NY, USA, Jul. 2020, pp. 76–89. [5] R. Ben Basat, S. Ramanathan, Y. Li, G. Antichi, M. Yu, and M. Mitzenmacher, “PINT: Probabilistic In-band Network Telemetry,” in Proceedings of the Annual conference of the ACM Special Interest Group on Data Communication on the applications, technologies, architectures, and protocols for computer communication, New York, NY, USA, Jul. 2020, pp. 662–680. [6] Y. Zhou et al., “Flow Event Telemetry on Programmable Data Plane,” in Proceedings of the Annual conference of the ACM Special Interest Group on Data Communication on the applications, technologies, architectures, and protocols for computer communication, New York, NY, USA, Jul. 2020, pp. 76–89. [7] H. Herodotou, B. Ding, S. Balakrishnan, G. Outhred, and P. Fitter, “Scalable near real-time failure localization of data center networks,” in Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining, New York, NY, USA, Aug. 2014, pp. 1689–1698.