可延展之開關式內連網路

Abstract

在此篇論文中, 我們提出了一個用在平行計算機系統的內部連接網路,此 網路叫做可延展之開關式內連網路. 可延展之開關式內連網路是由同樣的 基本模組所構成, 而每一模組包含一個開關及四個處理機. 這開關使得可 延展之開關式內連網路能模擬多種靜態及動態的內部連結網路.此篇論文 提出逐層建構此種可延展之開關式內連網路的方法, 並和超立方體內連網 路比較, 以凸顯其結構上的特性. 我們也試著將一些靜態的內連網路如 樹, 超立方體等崁在可延展之開關式內連網路中, 來說明它模擬靜態網路 的特性. 此外我們又讓可延展之開關式內連網路去模擬廣義立方體多層階 資料傳輸網路, 來說明它模擬動態網路的特性. 最後,我們更進一步提出 建構各種不同的可延展之開關式內連網路之可能性並討論之. A new class of modular interconnection networks is proposed for the hierarchical construction of massively parallel computer systems. These networks are called hyper-switching nets. The basic building block consists of four nodes and a switch. The switching elements make it possible for hyper-switching nets to embed many useful static and dynamic topologies. A recursive procedure of constructing hyper-switchings is presented in the thesis. The architectural properties of these networks are analyzed and compared with those of hypercubes. Static topolgies that can be embeded in hyper-switching nets include trees and hypercubes, etc. The generalized cube networks are simulated to illustrate the capability of the hyper-switching nets to embed dynamic networks. Some variations of the hyper- switching nets are also explored.