[DGD] Mutli-CPU Support?

[Felix A. Croes]

"Kevin N. Carpenter" <kevinc at seaplace.org> wrote:

> My concept is for several machines to dynamically carry the load 
> associated with different parts of the same mud.  Each part would be 
> dynamic down to its existance, with parts coming and going from the mud 
> on a slow timeframe (think hours to months, but without human 
> intervention).  Each part would be unique via its detailed definition, 
> but would, of course, require some common execution core.
>
> The good news is that a group of students have taken augmenting 
> OpenMosix with shared memory support on as their thesis project.  Of 
> course, shared memory over any network is going to be slow, so I hope 
> Felix's implementation utilizes shared memory lightly.

I think that you are unrealistically hopeful about the possibility of
combining the benefits of two very different projects, just because
both employ the term "multi-processor".  DGD/MP is essentially a
multi-processor LPC operating system, in competition with, rather than
as complement of, OpenMosix.

I am doing this for reasons very different from the ones that motivate
you: I care only about speed.  Therefore, having shared memory to me
looks like a very superior setup, indeed.


>[...]
> On a vaguely related note:  Memory is now cheaper than disk was not that 
> many years ago.  I'm curious if the DGD swapping algorithm can be 
> throttled, such that the mud won't swap unless resource utilization 
> exceeds some set thresehold.  Hmmm, I guess thats what the "cache_size" 
> startup parameter does, doesn't it?

Correct.  As long as your working set is smaller than the cache, there
will be very little actual disk access (note that DGD still counts an
object in the cache as swapped out, per the status() swap statistics).

Regards,
Dworkin
_________________________________________________________________
List config page:  http://list.imaginary.com/mailman/listinfo/dgd