Introduction

The problem

In a cluster of workstations or the Grid, each node has a certain amount of resources that can only be used from that node, if nothing is done. This restriction limits the power of such systems. The ideal scenario would one where all these resources (memory, disks, CPU, etc. )  become available from any node regardless of their physical attachment. Furthermore, this sharing would be perfect if it were done in a transparent way to users.

The solution

The proposed solution for this problem consists of implementing a middle-ware layer that glues togther all operating systems on all nodes offering a unified access to system resources. We call this a single-system image.

The benefits

The most important advantages of a system with a single-system image can be summarizes as the following:

• It frees the end-user from having to know where an application will run
• It frees the operator from having to know where a resource is located
• It greatly simplifies system management
• It reduces the risk of operator errors, with the result that end-users see improved reliability and higher availability of the system
• It provides independent message communication
• It allows the usage of any resource in the system in a transparent way from its physical location

The layers implementing it

The effect of a single-system image can be offered by one or more of the following layers

• Hardware layer
  Systems such as digital memory channel and hardware DSM offer single-system image at hardware level and allow the user to view a system as a shared-memory one.
   
• Gluing layer (operating system)
  At this layer, the gluing software can offer any object managed by any operating system as a global one to all applications. At this level we can build a single file system that manages all disks, or implement software DSM. OS kernels supporting single-system image include SCO UnixWare and Sun Solaris-MC.
   
• Applications and subsystem layer
  Finally, this kind of image can be supported by applications and subsystems. The application level single-system image is the highest and, in a sense, most important because this is what the end-user sees. At these level we can offer GUI based tools offering a single window representing all the resources or services.

The key services for a single-system image system

• Single port entry
• Single file hierarchy
• Single I/O space
• Single point of management and control
• Single virtual networking
• Single memory space
• Single job-management system
• Single user interface
• Single process space
• Single system disk

Related research links
(This list is by no means comprehensive. If you know of any other interesting link, please send it to me)

Academic Projects

• Snowflake Distributed System (Dartmouth College)
• Hive OS (Stanford University)
• GLUnix (NOW project) (UC Berkeley)
• Mosix (The Hebrew University of Jerusalem)
• OpenMosix
• Tornado (University of Toronto)
• Sprite (University of California, Berkeley)
• Nomad (COPPE Systems Engineering, Federal University of Rio de Janeiro)
• cJVM (IBM research - Haifa research laboratory-)
• Kerrighed (IRISA /INRIA - PARIS project)

Commercial products

• Memory Channel
• SCO UnixWare NonStop Clusters
• Solaris MC
• Single System Image Clusters for Linux from Compaq
• OpenVMS

Coordinator: Toni Cortes

Postal mail:
Departament d'Arquitectura de Computadors
Universitat Politècnica de Catalunya
Campus Nord - C6-118
C/ Jordi Girona 1-3
E-08034 Barcelona
Spain

E-mail: toni@ac.upc.edu or toni.cortes@bsc.es

Phone: +34 93 4016950
Fax: +34 93 4017055

This page is maintained by Toni Cortes. If you have any comments, please send him a message.

Disclaimer