You could always use a cluster file system like Red Hat GFS, Lustre, or Ceph to do something like this (at the most primitive level, you could even just use NFS or even set the disks up as iSCSI targets). However, if you go this route storage traffic on your cluster LAN can and probably will interfere with compute traffic. Plus, you will probably have trouble re-sharing disks that are themselves mounted remotely to the head node. In my opinion, if at all possible you need to buy a higher capacity head node.
|