Put the package you want to add in:

/home/install/contrib./4.0.0/arch/RPMS

Create a new XML configuration file that will extend the current compute.xml configuration file:

# /home/install/site-profiles/4.0.0/nodes
# cp skeleton.xml extend-compute.xml

Inside the extend-compute.xml, add the package name with a <package> tag.

Insert any post installation scripting in the <post> section.

This is a truncated version of an extend-compute.xml script we use:

<?xml version="1.0" standalone="no"?>
<kickstart>

<!-- There may be as many packages as needed here. -->
<package> kernel-smp </package>
<package> kernel-smp-devel </package>
<package> kernel-sourcecode </package>
<package> topspin-ib-rhel4 </package>
<package> topspin-ib-mod-rhel4 </package>
<package> compat-libstdc++-33 </package>
<package> panfs </package>
<package> F2PY-fix </package>
<post>

<file name="/etc/profile.d/topspinvars.sh">
if ! echo ${LD_LIBRARY_PATH} | grep -q /usr/local/topspin/mpi/mpich/lib64;

then

LD_LIBRARY_PATH=/usr/local/topspin/mpi/mpich/lib64:${LD_LIBRARY_PATH}
fi
if ! echo ${LD_LIBRARY_PATH} | grep -q /usr/local/topspin/lib64 ; then
LD_LIBRARY_PATH=/usr/local/topspin/lib64:${LD_LIBRARY_PATH}
fi
</file>

<file name="/etc/profile.d/topspinvars.csh">
if ( /usr/local/topspin/mpi/mpich/lib64 !~ "${LD_LIBRARY_PATH}" ) then
set LD_LIBRARY_PATH = ( /usr/local/topspin/mpi/mpich/lib64$LD_LIBRARY_PATH )
endif
if ( /usr/local/topspin/lib64 !~ "${LD_LIBRARY_PATH}" ) then
set LD_LIBRARY_PATH = ( /usr/local/topspin/lib64 $LD_LIBRARY_PATH )
endif
</file>

<!-- Setup for Portland Group Compilers -->
<file name="/etc/profile.d/pgi-path.sh">
LM_LICENSE_FILE=/space/apps/pgi/license.dat
export LM_LICENSE_FILE
export PGI=/space/apps/pgi
if ! echo ${PATH} | grep -q /space/apps/pgi/linux86-64/6.0/bin ;

then

PATH=/space/apps/pgi/linux86-64/6.0/bin:${PATH}
fi
</file>

<file name="/etc/profile.d/pgi-path.csh">
setenv PGI /space/apps/pgi
setenv LM_LICENSE_FILE $PGI/license.dat
if ( /space/apps/pgi/linux86-64/6.0/bin !~ "${path}" ) then
set path = ( /space/apps/pgi/linux86-64/6.0/bin $path )
endif
</file>

</kickstart>

To apply the customized configuration scripts to the compute nodes, rebuild the distribution:

# cd /home/install
# rocks-dist dist

Time for a break. ***time 2:30:00

Switch Configurations
- Setup IP information
- Enable "fast link" on all ports
- Link aggregation (four 1GB aggregated links between switches, four 1GB aggregated link for the Panasas Installation)
- Setup configuration on Topspin 270 switch

Insert the Compute Nodes
- Invoke 'insert-ethers' on the front-nd node and select 'compute node' for the appliance type (insert-ethers is used to discover compute nodes and ensure they get the proper configuration)

I chose to make modifications to each compute node for remote management access for remote boot/power cycle, etc. This added two minutes to each node's install time. I also like to make sure each node is discovered in the order in which I power them on (left to right, bottom to top in the racks). This process takes more time, but it saves a huge amount of time when tracking down hardware failures. It also makes labeling them following the compute node installation easy, with the utility provided by Rocks for creating labels with identification information. (This little tool really shows that the distribution was developed by people who manage clusters.) 48 nodes x 3.5 minutes per node = 2 hours 48 minutes for this process due to my requirements. It could be far less time depending on your needs.

Time for a break. ***time 5:40:00***

Time for Panasas Integration
The Panasas Storage Cluster arrived in three boxes on one pallet. From the time I clipped the first band on the pallet to having the system fully operational was only 1 hour 55 minutes. Here's how the process went.

First off, we unpacked the equipment and identified all components and parts. I found a location in the rack, set up the mounting hardware then mounted the chassis. I loaded the chassis with one DirectorBlade and 10 StorageBlades, network, battery and power supply modules, and connected four 1GB Ethernet links to the link aggregation point on the switch.

Now it's time to configure the software side of it.

- Connect a console cable from my laptop to the serial port on the DirectorBlade
- Log in as admin
- Configure the system via the pancli (a modified Panasas command line interface on the Unix-based operating system) with IP information

I was then able to connect to the system via the Web GUI to complete some additional configuration:

- Customer ID
- System name (required for global file system configuration)
- IP range for DHCP pool for Director Blade and StorageBlades
- Timezone

This minimum configuration is all that's required for the system to function as a basic network attached storage (NAS) system. Additional modifications can include DNS, NIS, NTP, NFS, CIFS, and SMTP servers (notifications). You can also modify volume and quota information. There are some modifications to make to the useradd scripts based on setting choices and auto-mount options if you want user home directories located on the Panasas Storage Cluster, one way in which we use the unit.