JackNeely jjneely@pams.ncsu.edu

The next generation of software solutions for high performance computing offered by PAMS Computer Operations is a Red Hat Linux based Beowulf distribution called NPACI Rocks. We are phasing out the very old NCSU Scyld distribution in favor of this newer and more powerful software. The goal of this short document is to provide some instructions about installing and realmifying the Rocks distribution. The new software contains many features. Many of which are much more advanced and mature when compared to the NCSU Scyld Beowulf distribution. The software contains a web based monitoring tool, MPICH, PVM, PVFS, OpenPBS, the Sun Grid Engine, etc. One of the major goals of this project is easy maintenance, and it is pretty easy to install and maintain.

Here I will point you to the NPACI Rocks Documentation. It contains all the instructions for getting the software, installing it, and start computing with it. Note that it is easier to complete the rest of this document after you have installed the front end machine and before you have installed any of the slave nodes.

Realmification of the Rocks distribution is pretty simple. All you need to do is to install the ncsu-beowulf package on the cluster. This package will configure kerberos to interface with the kerberos authentication system here at NCSU. It modifies one of the PAM configuration files. Finally, it installs the allowuser and rmuser commands. Which will allow you to let NCSU realm IDs log into the cluster. Both command accept the --help argument to display how they are to be used. The latest version of the ncsu-beowulf package is availible on kickstart.linux.ncsu.edu in the /pub/contrib/jjneely/NPACI_Rocks directory.

By now you should be able to install your slave nodes and have a working system. However, you may want to do some other configuration before hand. The configuration you do is all dependent on how you would like to use your cluster. The following are some pointers to set up a cluster similar to how PAMS plans to use this software.

You will need to create the file /etc/USESGE to switch to the Sun Grid Engine from the default batch system (OpenPBS). You should do this before you install your slave nodes. This way the insert-ethers program will be able to do some of the configuration for SGE for you. Rocks version 2.3 had some bugs in the SGE packages so you will want to at least use Rocks 2.3.1.