It seems to me that the data is much more strongly bound together row-wise than column-wise. People who use this will probably want to access information about logical volumes, i.e., they will want to treat LV NAME as a primary key. So I would recommend the hash structure as most natural:

$value = $hash->{$lv_name}{$attribute};
[download]

Presenting the data in parallel arrays requires one to think about an abstract index to synchronize data extraction, which is less intuitive than the LV NAME.