From photons to big-data applications: terminating terabits.
From photons to big-data applications: terminating terabits.
复制标题
从光子到大数据应用：终止特拉比特。
DOI：
10.1098/rsta.2014.0445
复制
发表时间：
2016-03-06
期刊：
影响因子：
--
通讯作者：
Crowcroft JA
中科院分区：
文献类型：
--
作者：
Zilberman N;Moore AW;Crowcroft JA
计算机体系结构已经进入了一个分水岭，因为用户应用程序生成的网络数据量超过了任何单个计算机终端系统的数据处理能力。如果联网数据的数量与每个系统的数据处理能力之间的差距越来越大，就不可能扩展现有的计算机系统。尽管如此，对任务多样性和任务复杂性的需求仍然有增无减。联网的计算机系统为新的应用程序的开发提供了一个肥沃的环境。随着联网的计算机系统变得类似于基础设施，对容量和能力的增长的任何限制成为所有计算机用户关注的重要约束。考虑到一个联网的计算机系统能够处理每秒太比特，作为可扩展性的基准，我们批判的最先进的商品计算，并提出了一个批发重新考虑在计算机体系结构的设计和随之而来的生态系统。我们的建议旨在降低成本，节省电力和提高性能的多尺度方法，具有潜在的应用从纳米级到数据中心规模的计算机。
Computer architectures have entered a watershed as the quantity of network data generated by user applications exceeds the data-processing capacity of any individual computer end-system. It will become impossible to scale existing computer systems while a gap grows between the quantity of networked data and the capacity for per system data processing. Despite this, the growth in demand in both task variety and task complexity continues unabated. Networked computer systems provide a fertile environment in which new applications develop. As networked computer systems become akin to infrastructure, any limitation upon the growth in capacity and capabilities becomes an important constraint of concern to all computer users. Considering a networked computer system capable of processing terabits per second, as a benchmark for scalability, we critique the state of the art in commodity computing, and propose a wholesale reconsideration in the design of computer architectures and their attendant ecosystem. Our proposal seeks to reduce costs, save power and increase performance in a multi-scale approach that has potential application from nanoscale to data-centre-scale computers.