Have one YAML format for all areas.
All targets, units of racks, ports of KVM, PDU, network switches, are identified by numerical ID, so we can use it for identification. Also, since most of devices in racks have their hostname for network connection, it is possible to list by hostname with port identification on hardware side for identification.
For yaml format, only "units in racks" can have multiple unit IDs for one hardware, but others (e.g. ports of switches, PDU, KVM) not, also unit ID in rack will not be read often by software not like ports of switches which could be used for statistics display. So, it will not be harmful even though we have the same values for multiple keys when we set keys as IDs.

Format proposal (to be written in README.rst):

• top level as mapping consisted of "device" and "assign"
• "device" contains device information and is organized as mappings
• "assign" contains unit registration and is organized as sequence of mappings
• each mapping in "assign" is for one unit ID, and contain hostname as name.