In this lecture I will introduce several optimization problems involved in designing supercomputers, and we will discuss how to solve them. A future supercomputer is a massively distributed system that consists of some tens or hundreds of thousands of computing nodes connected with an interconnection network. As a processor alone is no longer getting faster, a supercomputer must exploit more number of processors to get faster, where the interconnection network becomes a bottleneck. We will focus on the design of the interconnection networks from the viewpoint of graph theory, and will discuss how to formalize them as optimization problems. Those optimization problems seem to be traditional (e.g. the degree/diameter problem, the graph embedding problem, the quadratic assignment problem, and the job scheduling problem); however, it is not very easy to solve them. Let's discuss together to find a smarter way to design future supercomputers!
Ikki Fujiwara received the BE and ME degrees from Tokyo Institute of Technology, Tokyo, Japan, in 2002 and 2004, respectively. He worked for Hitachi Ltd., Ibaraki, Japan, developing railway traffic management systems from 2004 through 2008. He received the PhD degree from the Graduate University for Advanced Studies (SOKENDAI), Tokyo, Japan, in 2012. He is currently a Project Assistant Professor in the Information Systems Architecture Research Division, National Institute of Informatics, Tokyo, Japan. His research interests include the areas of high-performance computing, networking and optimization.
