Recently, Professor Shao Ziyu’s research group made a breakthrough in fundamental theory of networking, solving an open problem: how to systematically design effective large-scale network topology? Their article “Systematic Topology Design for Large-Scale Networks: A Unified Framework” was accepted by IEEE International Conference on Computer Communications (IEEE INFOCOM 2020).
In the field of networking, there has long been an open problem for a long time: how to design network topology systematically for various performance requirements with a top-down approach. Existing work usually adopts a bottom-up approach with heuristics, which can only be applied to some specific scenarios, and it is very challenging to be extended to more general scenarios.
To solve this problem, Professor Shao and his group spent two years on a general design framework proposal. The solution adopts a combination of reverse engineering and forward engineering. In fact, there is a long-held belief that empirically excellent engineering design should have a corresponding mathematical principle and an implicit math problem. Finding such math principle and problems lead to a systematic understanding of the engineering design.
More specifically, the first key step is to reverse engineer the classical Fat-Tree topology. The math principle behind such design is revealed, and a more general topology design framework is proposed based on such principle. The researchers found that nearly all important network topologies in data center networking (cloud computing) and supercomputing can be rediscovered from the framework. Furthermore, with such framework, many forward engineering works have been done and several new topologies have been proposed for various fields including data center network, supercomputing, federated learning, intelligent internet of things, and integrated circuit design.
For this work, students Chang Yijia, Huang Xi and Deng Longxiulin in Professor Shao’s group are student authors and Professor Shao Ziyu is the corresponding author, with ShanghaiTech University as the first responsible institution. The work was supported by start-up funding from ShanghaiTech University, National Natural Science Foundation of China, and Natural Science Foundation of Shanghai.
Example of New Network Topology I
Example of New Network Topology II
Example of New Network Topology III