Grid.org

Well if you haven't already checked it out please go ahead and have a look on http://groups.grid.org/content/unicluster for the UniCluster 4.0 Test Drive.

UniCluster 4.0 is a big change from our previous releases of UniCluster.  We went back to the drawing board and 'put our heads together' to figure out 'what is needed for a medium sized HPC cluster?'.  UniCluster 4.0 creates the whole cluster all you need is the hardware and CentOS 5 or Oracle Enterprise Linux (the software works on RHEL as well but we didn't focus on that for the Test Drive.)

Anyhow enough talk about UniCluster 4.0 Test drive....does it really work?

 I have tried it on several vmware fusion virtual machines on my Mac Powerbook and it works fine!  It does take some time to get a 'clean' version of CentOS or Oracle Linux installed but once that is done the UniCluster installation is pretty easy.   I liked the simplicity of clicking on the 'repo rpm' that configures the yum repository for UniCluster and then all I did was a 'yum install unicluster' to download all of the base packages for UniCluster onto my new cluster 'installer node'.

Once the base packages were installed I sourced the Unicluster environment and ran the setup script. This part took some time as well since the installer asked for the CentOS media (it uses it to create the repository for the compute nodes) once the media is read and a repo is created the installer creates a disk and diskless image for the cluster.  After all that the setup finishes and the installer node is done.

 So after installing I fired up the web browser and pointed it to http://localhost.  That gave me a cluster homepage with documentation and installed 'kits'.

There a quite a few Kits available on the yum repository, you can have a look at the list by running the command:

# yum list unicluster-kit-*

I installed the SGE kit with the following command:

# yum install unicluster-kit-SGE

Then configured it by running:

# /opt/kusu/sbin/install-kit-sge

Once the Kit is installed it didn't do anything until I realized that it was not supposed to do anything until the kit is associated with a group of nodes.  To associate a kit with nodes I ran the node group editor 'ngedit' which is a menu driven tool for building clusters and associated the 'installer node' group with the SGE main component.  Once I did that and re-sync'd the node (it takes a while to do this and UniCluster displays a message saying it is busy...) I had SGE installed on the installer node in the cluster.

SGE was installed but I still had to source the environment or logout and log back in to run the SGE commands and submit jobs into the cluster....SGE takes some time to get used to but once you figure it out it is pretty easy to submit jobs to the system.

Overall the install was pretty easy...there are some concepts that take some time getting used to but after you do it once it is no problem.

The only thing I messed up was I did not sync time properly among the nodes in the cluster by running NTP....once I fixed that the system started to behave better.

Well that's all for now. I'll post more later.