Distributed Systems Laboratory
Argonne National Laboratory
University of Chicago

DSL DataGrid Cluster User's Guide

Table of contents:


Overview

The Datagrid cluster is a 20-node cluster designed for collaborative use. It will run current versions of Globus so that people can test against a known environment.

The resources of the Datagrid are as follows:

  • 20 nodes with dual 850MHz PIII processors and 512MB memory each (dg0n1.mcs.anl.gov -> dg0n20.mcs.anl.gov)
  • 2TB of shared storage using the PVFS filesystem, mounted at /shared.
  • Globus 2.0, located in /soft/pub/packages/globus/globus-current, with a PBS gatekeeper running on dg0cm.mcs.anl.gov. We will trust the Globus CA, the DOE SG CA, and others.

Getting access

To request an account on the Datagrid, please visit the MCS accounts webpage and request access to the DSL DataGrid Cluster.

Once you have an account, please email dsl-support@mcs.anl.gov to be added to the list of authorized users of the PBS scheduler. In the email you will need to include:

  • Your local account name
  • A project name (e.g., Fusion Grid, ESG, DOE SG, etc.) for accounting purposes.
  • The DN from your certificate you want entered in the grid-mapfile
This information will be verified and you will be notified when you are authorized.

Using the Cluster

Hostnames

The cluster consists of 20 compute nodes named dg0n[1-20].mcs.anl.gov (e.g. dg0n7.mcs.anl.gov) and a compute manager (dg0cm.mcs.anl.gov).

Interactive Access

SSH may be used to log into each of the compute nodes. The compute manager does not offer interactive access.

Filesystem

The cluster has a 2TB shared filesystem in /shared. Please Note! All disk storage should be regarded as volatile. This system is designed to provide a test resource for Globus applications, and is not a production research resource. No unique data should be stored on this cluster.

GridFTP

Each of the compute nodes runs a GridFTP server on the standard port.

Job Submission

The compute manager allows submitting of jobs to the PBS scheduler running on the cluster. To submit jobs to the PBS scheduler use the dg0cm.mcs.anl.gov node: XXX Insert example.

The compute nodes each have a Gatekeeper and JobManager supporting fork/exec.

MDS

There is a GIIS on dg0cm under "mds-vo-name=anl-dg, o=grid".


Please send comments regarding this website to dsl-webmaster@mcs.anl.gov.