discussion:Where do we go from here?

Andy Thornton A.J.Thornton@durham.ac.uk
Tue, 8 Nov 1994 13:52:25 +0000 (GMT)



On Mon, 7 Nov 1994, Raul Deluth Miller wrote:

> Raul Miller:
> . > also: machine abstraction suitable for describing system
> . > capabilities [for estimation and planning purposes]
> 
> Mike Prince:
> . I'm a little fuzzy here...  Do you mean some kind of benchmarks for
> . different capabilities, or the ability to do benchmarks?
> 
Raul Miller:

> 
> Obviously, we're not going to even have good estimates all the time
> (on what kind of resources we're going to need), but we should be able
> to act smart with decent estimates.
> 

Do we need a set of requirements for each object?  For example:

- hardware floating support prefered
- sound capability essential
- indication of memory useage
- with what other objects will commnuication be most frequent - this is 
  to help prevent closely connected objects from being migrated to far 
  from each other.

One solution to the last problem is that one criterion for the migration 
algorithm should be to keep the messaging path between closely 
interlinked objects to a minimum - ie to have a minimum cost spanning 
tree for the communication paths between an object and all it's 
users/clients.

Andy