SEC: object security, LLL,DIS,HLL

Mike Prince
Tue, 1 Nov 1994 15:32:10 -0800 (PST)

On Tue, 1 Nov 1994, Raul Deluth Miller wrote:

> Mike Prince:
> . A more fundamental question is how many objects should you be able
> . to have?  It's a low level issue that percolates up to the top.  If
> . my program assumes it can have a zillion objects, starts running on
> . a machine that can't even address a zillion objects than we have a
> . problem.
> Well one thing: if there are going to be different sized machines in
> the system the answer to this question will be different for different
> locations.  This sort of question (what resources does this program
> need) needs to be addressed in the implementation of the migration
> algorithm.
> Also note that someone with a bunch of small machines is going to want
> different behavior than someone with a few big machines.

Two things I'm touching on.

One is some sort of a priori knowledge about an application given to the 
OS would be "nice".  For instance, how much memory might it use, how much 
would it use other tools (how fast/slow could they respond, how often), 
how many agents it might spawn, etc.

I'm not saying we'd have to provide this information (or even that we 
could), but lets think about whos role it would be, how helpful would it 
be, etc.

The second issue goes down to the hardware independence thing.  If we 
standardize the way we store data (one of my big If's) then a program 
might have an integer to store a handle to an object.  That would be in a 
standard stack and not be touched when migrating between machines.  
Because it can't be made "bigger" than whatever size we choose, we are 
stuck with.