Design and Implementation of Clusters with Single Process Space

단일 프로세스 공간을 제공하는 클러스터 시스템의 설계 및 구현

  • Park, Min (Division of Computer Science, Department of Electrical Engineering and Computer Science Korea Advanced Institute of Science and Technology) ;
  • Lee, Daewoo (Division of Computer Science, Department of Electrical Engineering and Computer Science Korea Advanced Institute of Science and Technolog) ;
  • Park, Dong-Gun (Division of Computer Science, Department of Electrical Engineering and Computer Science Korea Advanced Institute of Science and Technolog) ;
  • JungLok yu (Division of Computer Science, Department of Electrical Engineering and Computer Science Korea Advanced Institute of Science and Technolog) ;
  • Maeng, Seung-Ryoul (Division of Computer Science, Department of Electrical Engineering and Computer Science Korea Advanced Institute of Science and Technology)
  • Published : 2004.04.01

Abstract

Single system image(SSI) have been the mainstay high-performance computing for many years. SSI requires the integration and aggregation of all types of resources in a cluster to present a single interface to users. In this paper, we describe a cluster computing architecture with the concept of single process space(SPS) where all processes share a uniform process identification scheme. With SPS, a process on any node can create child process on the same or different node or communicate with any other process on a remote node, as if they are on a single node. For this purpose, SPS is built with the support of unique cluster-wide pid, signal forwarding, and remote fork. We propose a novel design of SPS cluster which addresses the scalability and flexibility problem of traditional clusterwidely unique pid implementation by using blocked pid assignment. We have implemented this new design of SPS cluster, and we demonstrate its performance by comparing it to Beowulf distributed process space. Benchmark performance results show that our design of SPS cluster realized both scalability and flexibility that are essential to building SPS cluster.

Keywords