During whole history of existance of computer, there was three generation of storage systems. The first has no even own terminology. Because absence of it, we will use terminology of second generation.
Distinctive particularity of systems of first generation was command 'move file', shifted whole file, and widened free space between files for writing of a new file.
Distinctive particularity of systems of second generation was cutting of files into pieces of the same size, putting of pieces into free spaces, and listing of these places in special stealthy file (FAT).
Systems of third generation was based on the fact, that except scientific calculations (with arrays), structures are used in all other cases, and query for structures is more usable and more functionally, than treatment with whole file. Besides, these systems had one more advantage. Memory, allocated by operational system for cashing of hard disk, is limited. And cashing of structures of all types at once in one cash (what would be equivalent to cashing of all files at once in one cash in systems of second generation) is more optimal, than cashing of each file in separate equal part of memory. Certainly, that all will work after moving of structures onto new hard disk, structures have to refer to each other not by physical address, not by serial number in chain of structures of identical type, but by number, given in order of birth of structures (of course, index also have to be supported by operational system to discharge apply program. Besides this, nested schemas have to be supported - p.176-178 of pdf-document).
Unfortunately, inventor of system of third generation (E.F. Codd) had no access to kernel of any OS, and cannot build it into there. And what is just more sad, he merged storage system with one of many possible machine of inference, by what he have confused all next generations of developers. Now DBMS consist of part of OS, and of machine of inference for operator 'join' - and neither developers of OS, nor developers of machines of inference cannot separate these two facts in own mind.
I bumped with it, when i intended machine of inference, differ from 'join' (algorith of inference - and even algorithms - were already known). And i was forced to attach my machine of inference to DBMS instead of OS (p.209-271 of pdf-document), because storage system did not implemented in OS.
Today dimness of minds is observed in our industry - something similar to dance fever or tulip fever in medieval Europe. Rollback from structured world-outlook to file world-outlook has occurred, marker of which is saving of XML.
I want to draw attention of community, that generations of storage systems do not changing because some factor, and to call, and to incline to build storage systems of 3-rd generation into operational systems.
Dima Turin, dmitryturin@yandex.ru