OFF: Extensible File System

Andy Thornton
Tue, 8 Nov 1994 13:29:03 +0000 (GMT)

On Mon, 7 Nov 1994, Mike Prince wrote:

> On Mon, 7 Nov 1994, Cybernaut wrote:
> >     I believe that by handling objects files as just another extended dirent
> > tag, the system can easily discern one object from another and know what
> > object does what without ever having to know where that object is stored or
> > anything about the internal stucture of the object files.
> Here's another version.  Forget everything you've learning about file 
> systems.  All objects are "floating around" in this virtual machine 
> (the set of all computers).  There are several rings around groups of 
> these objects.  All objects that just happen to be in a workspace (CPU), 
> all objects in a domain (group of CPU's), and finally the set of all of them.
> Objects would also contain static refernces to other objects, the links 
> of program flow.  These links could also be used to group objects or 
> navigate the relationships between them.
> Objects would have the ability to respond to requests.  Ask an object for 
> an icon of itself, or a dialog box to talk to it, etc.
> Just a note, what I'm describing are actually my Tool Boxes, though 
> people like to hear the term objects.  Each tool box could also contain 
> sub-objects which are manipulated by the code within the Tool box.  These 
> sub-objects could be the traditional OO objects with inheritance, etc.
> My main point, is that we could get completely away from the file system 
> hierarchy to a more WWW type system of managing data.  All data is free 
> floating.  Traverse links to get where you want.
> My $.02
> Mike
I think that this is the way to go but we will still need a framework in 
which to place out objects - is they will still need names of some sort - 
ie URL in the WWW system.

How do we cope with object migration?  Do we need a central database of 
object location?  Does this not threaten the system if the machine 
holding this database goes down?  Do we have to resort to polling all the 
machines to find where a particular object is currently residing.

>From what I can see the aim is to make a much more dynamic system for 
information storage but I think we need to balance this against the 
understandability of a static system - there must be a reasonable compromise.