[sc34wg3] N0391-0394: New SAM/XTM documents

Robert Barta sc34wg3@isotopicmaps.org
Tue, 22 Apr 2003 12:01:02 +1000


On Mon, Apr 21, 2003 at 07:00:11PM -0400, Sam Hunting wrote:
> > > Ah, ok.   But then....why would you distribute 100.000 topics over
> > > 5 machines? 
> > 
> > Uh, I want to keep my Linux cluster busy :-). Seriously, the
> > application I am thinking about will need a cluster, so distributed
> > processing is a must.
> 
> So the fact of >100,000 sparsely connected topics does not create a
> requirement (in your view) for a cluster? I think that's what Jan was
> asking.

Sam,

I am not sure whether he was asking this.

Anyway, 10^5 topics can be held in main memory of a single
machine. Assuming a reasonable TMDBMS implementation.

Let's make a quick calculation:

  - 100000 topics, 3 assocs per topic
  - topic memory 200-400 bytes (?)
  - assoc memory < 100 bytes (?)


   topics ......     100000 x 400 +
   assocs ...... 3 x 100000 x 100
------------------------------------
                               70 MB
   + indexing factor 1.5
====================================
                            ~ 100 MB

So storing alone is not the problem, processing (querying) may,
depending what you want to do with it.

\rho

PS: I am - of course - not talking about implementing it in
    Jaaaaaaaaaaaavaaaaaaaaaaaaa ;-)