Re: [ng-spice] Rework 9 - configure problems


To ng-spice@ieee.ing.uniroma1.it
From "James Swonger" <jwsrh@hotmail.com>
Date Thu, 30 Mar 2000 07:51:20 PST
Delivered-To mailing list ng-spice@ieee.ing.uniroma1.it
Mailing-List contact ng-spice-help@ieee.ing.uniroma1.it; run by ezmlm
Reply-To ng-spice@ieee.ing.uniroma1.it

I'd concur that fixing failure modes is number one priority.
Memory manager robustness is key to being able to use SPICE
for designing complex analof circuits, or even relatively
simple ones that have complex behavior/environments (e.g.
a PWM with 1000 active and passive components, but which
might need to run for 100uS at a 10nS timestep (10K total
steps, or more). Some commercial simulators cannot deal
with data volume of this magnitude. A simulation that
bombs is not helpful ;(   A simulator that bombs easily
and repeatedly is soon discarded.

I would like to see this simulator become so robust, as
regards memory management and data size, that you could
run until your rawfile hits the disk quota and your
active data space (matrices?) hits the swap/temp space
limit.



> > - Where do we want to go ?
> >   This is still an open question. There was a short thread last week 
>about
> >   writing the API's for a simulation kernel to use as base for a new
> >   spice. Should we do this ? Can we do this ? I have spent some time
> >   in the netlib repository searching for algorithms and packages for
> >   sparse matrix handling and came out with a huge state of confusion.
> >   I am definitively a numeric-ignorant. Before spending time in other
> >   parts of the code, I think we should address the matrix package
> >   problem. Sparse seems still a good code (as written in numerical
> >   repositories), but it can improved for new deeply pipelined
> >   architecture.
>
>     I also does not know much about numerical problems, but I am not sure
>about what should be fixed first. I think that (if we stay with old spice) 
>we
>should rather fix memory leaks and other BUGs then thinking about better
>algorithms. I said this based on my experience with spice user interface, 
>so
>"simulation code" problems might be different, but this is my opinion.

______________________________________________________
Get Your Private, Free Email at http://www.hotmail.com


Partial thread listing: