Diskless Debian Cluster

In our group we use molecular dynamic simulations to research the dynamics of hydrogen bonding molecules, most notably water, in confinement and on surfaces. In order to run this simulations we set up a small computing cluster. With now 20 nodes it was practical to resort to diskless nodes. Firstly, this is due to cluster management , as we need now only one master image for all nodes. Secondly, there are no more any hard disks in the nodes which may fail. This diskless setup runs now very smoothly, if there is any hiccup on a node it is usually solved by just rebooting the offending node.

The primary setup is done with a bash-script and the whole process is described in the manual. Both are available on bitbucket repository.

The current plan is to move to a single iSCSI disk per each node. The whole nfsroot and aufs layered on top creates problems when we want to use the nodes as storage nodes for ceph or gluster distributed storage file systems, as they need to store data per client on the host. How this is done with the current setup is described at the end of the [diskless manual](https://bitbucket.org/mrosenstihl/debian-diskless-cluster, it is possible but feels like a hack.

FAI in combination with salt to create iSCSI boot disks looks quite promising. Preliminary tests with two nodes show that it works in principle. This means that all the disks in the nodes will be available for storage.