This website requires JavaScript.

Mars: Near-Optimal Throughput with Shallow Buffers in Reconfigurable Datacenter Networks

Vamsi AddankiChen AvinStefan Schmid
Apr 2022
摘要
The performance of large-scale computing systems often critically depends onhigh-performance communication networks. Dynamically reconfigurable topologies,e.g., based on optical circuit switches, are emerging as an innovative newtechnology to deal with the explosive growth of datacenter traffic.Specifically, periodic reconfigurable datacenter networks (RDCNs) such asRotorNet (SIGCOMM 2017), Opera (NSDI 2020) and Sirius (SIGCOMM 2020) have beenshown to provide high throughput, by emulating a complete graph through fastperiodic circuit switch scheduling. However, to achieve such a high throughput, existing reconfigurable networkdesigns pay a high price: in terms of potentially high delays, but also, as weshow as a first contribution in this paper, in terms of the high bufferrequirements. In particular, we show that under buffer constraints, emulatingthe high-throughput complete-graph is infeasible at scale, and we uncover aspectrum of unvisited and attractive alternative RDCNs, which emulate regulargraphs of lower node degree. We present Mars, a periodic reconfigurable topology which emulates a$d$-regular graph with near-optimal throughput. In particular, wesystematically analyze how the degree $d$ can be optimized for throughput giventhe available buffer and delay tolerance of the datacenter.
展开全部
图表提取

暂无人提供速读十问回答

论文十问由沈向洋博士提出,鼓励大家带着这十个问题去阅读论文,用有用的信息构建认知模型。写出自己的十问回答,还有机会在当前页面展示哦。

Q1论文试图解决什么问题?
Q2这是否是一个新的问题?
Q3这篇文章要验证一个什么科学假设?
0
被引用
笔记
问答