next up previous contents
Next: Cluster nodes Up: Centralized Cluster Installation and Previous: Centralized Cluster Installation and   Contents

The DULUG linux installation site

The key to scalable linux installation on campus comes down to several things:

In a nutshell, Red Hat's current distribution(s) are mirrored to install.dulug.duke.edu and augmented with rpm (re)packaging of Duke's site license software and any other software found to be useful that is missing from the primary Red Hat distribution. Certain tools and configuration features are customized for the Duke environment. install.dulug.duke.edu is accessible only to duke.edu addresses.

From install.dulug.duke.edu, any member of the University community can install a working version of linux on nearly any networked computer, over the network, for free, in a matter of ten or twenty minutes. No particular expertise is required other than the ability to follow simple directions, although installing a computer located in a departmental LAN is best done with the support and cooperation of the LAN manager.

Kickstart is a Red Hat linux feature that permits the installation of a workstation or cluster node to be ``scripted'' from a straightforward template. Work done once developing a single LAN-specific kickstart script can then be used many times to install workstations, servers, cluster nodes. This is the installation tool of choice for LAN managers on campus that run any significant number of linux systems in their departmental LAN. With kickstart, to install a system (or reinstall a system) one simply boots it, either from floppy, from the local hard disk, or (using a PXE-enabled ethernet card) from the network. Ten minutes later (or less, depending on bandwidth to and load on install.dulug.duke.edu) the node reboots itself into a fully installed and configured state.

Yum is a tool authored by Seth Vidal (the systems manager of the Duke Physics Department, who also maintains install.dulug.duke.edu) that fully automates various aspects of the maintenance of workstations, servers, and nodes after their original install. It is automatically installed and configured to run nightly on any Duke system installed from install.dulug.duke.edu that does not explicitly choose to exclude it. With yum, updating every linux-based system on campus can be as simple as dropping the updated package in the primary archive on install.dulug.duke.edu - yum will automatically install the updates from a nightly cron script.

This is crucial feature as it significantly increases the security level of every linux system on campus equipped with the feature - security updates are distributed within as little as 24 hours of the discovery of a problem, even onto systems belonging to totally system-ignorant individuals who would otherwise never hear of the problem or think to update their systems. Yum also greatly simplifies all the aspects of rpm package management for systems or LAN managers as it forms a consistent interface to the entire package collection on install.dulug.duke.edu (and/or other yum repositories set up in departmental LANs with specific site-local packages). Installing, removing, listing and otherwise maintaining packages is reduced to a single, simple command with a few options. With it and a simple script, a LAN manager can (for example) install a new software package on every system in their LAN with a single command typed once at their own desktop.

As long as the primary job of maintaining install.dulug.duke.edu and yum continues to be done by Seth Vidal (with the help of various other linux experts on campus, but his primary responsibility), one can see that his labor enables linux installation and software management to be done at the practical limit of efficiency (and amazingly close to the theoretical limit in efficiency). It isn't quite true to say that to install a new, PXE-equipped system in a campus LAN one has but to turn it on, but only a small amount of work at the LAN level (developing a kickstart file, setting up PXE and dhcp, writing the LAN-specific post install scripts that make the system a trusted member of this LAN and e.g. mount the correct server directories rather than a standalone untrusted host) makes it true.


next up previous contents
Next: Cluster nodes Up: Centralized Cluster Installation and Previous: Centralized Cluster Installation and   Contents
Robert G. Brown 2003-06-02