We are currently running an OpenNebula cluster that has an LVM based datastore that is located on a SAN device (Dell Unity). The SAN device has two independent storage processors and is configured to our nodes via a fiber-channel switch with multiple paths to each OpenNebula node.
The current implementation of the OpenNebula LVM datastore creates logical volumes on volume groups that are named when datastores are created manually, like this:
The logical volumes that OpenNebula creates on this group are named like this:
Our virtual group for the datastore currently has two physical LUNs that are on different storage processors on the SAN device. However, because the logical volumes are created as linear volumes by default, they only fill the first physical device or LUN. We would like to transform the linear logical volumes to striping or raid0 types. We know how this conversion would be done using the LVM tools. But are there any pitfalls that should be taken into consideration? Has anyone attempted something similar?