DPCell: Constructing Novel Architectures of Data Center Networks on Dual-Port Servers

Autor: Yanming Shen, Heng Qi, Deshun Li, Keqiu Li
Rok vydání: 2021
Předmět:
Zdroj: IEEE Network. 35:206-212
ISSN: 1558-156X
0890-8044
DOI: 10.1109/mnet.011.1500102
Popis: To store and process big data, large-scale data centers are built to interconnect a huge number of servers, which is a challenging task due to the network scalability and performance. In this article, we study the underlying common principles behind the existing data center network architectures on dual-port servers. We find that the existing architectures hardly achieve both large scalability and bisection width simultaneously. As a result, we propose a novel fabric of switches (FoS), which is recursively constructed on the flexible division of switch ports by means of a complete graph. The novel FoS benefits data center networks by flexible design space and double-exponential scalability, as well as small diameter and large bisection width. DPCell is constructed by inserting a dual-port server between any two adjacent switches in the novel FoS, which inherits good properties of the embedded FoS. Comparison with other architectures shows DPCell can achieve both large scalability and bisection width with a small diameter. Evaluation results show that DPCell demonstrates good performance in average routing delay and average load on servers under different degrees of network traffic.
Databáze: OpenAIRE