it was also in illustra, stonebraker's attempt to commercialize postgres in the 90s. every table had meta columns defining bounds for validity for this version

it was also in illustra, stonebraker's attempt to commercialize postgres in the 90s. every table had meta columns defining bounds for validity for this version of a given row. an update copied the row, applied changes, updated validity timestamps on the old and new copies. aside from the disk space problem you mention, the other issue was that every update required a write to every index on the table, even if the indexed column hadn't been changed. makes simple updates much heavier than one might naively think...