In einer eMail vom 12.07.2008 02:57:44 Westeurop�che Normalzeit schreibt
bill@herrin.us:
On Fri,
Jul 11, 2008 at 7:05 PM, <HeinerHummel@aol.com> wrote: > My
understanding of what you want is this: forwarding from C to B is not >
allowed. Hence the packet from D to E must go like > this:
D--->C-->G-->F-->E and NOT
D-->C-->B-->F-->E. > My saying is: That can be respected.
But this is almost normal Dijkstra.Only > the following enhancements are
required: Build a graph, which consists of > these nodes A,..H and of
directed arrows. Between any two nodes according to > the black lines
there are two inversely directed arrows.Except between B >
and C. Here there is only one arrow which is from B to C, but not from C
to > B. > Node C runs Dijkstra with itself being the root, however
modified such that > selecting a predecessor node for any node presumes
that there is an arrow > from that predecessor node to that
"any"-node. So , while running Dijkstra, > C won't even see that B is a
neighbor/candidate.Hence the resulting shortest > path tree will not
include "from C to B" but only "from C to G".
Okay. I added what I
think you just described in red
at http://bill.herrin.us/network/geoag-h1.gif. Is that
correct?
Yes.
This
means that the forwarding information base at node G looks
like:
destination {left area}: send to F destination C: send to
C destination H: send to H destination D: send to C
And the FIB
at node C looks like:
destination {left area}: send to G destination
G: send to G destination D: send to D destination H: send to G
Is
that correct?
Yes.
So,
D--->C-->G-->F-->E happens for packets from D to E because
when the packet is at C, C understands that all of {left area} is
available via G and when the packet is at G, G understands that all of
{left area} is available via F.
Still correct.
Thus
geographic aggregation cut the FIB almost in half since C and G only have
to keep track of {left area} instead of keeping track of A, B. F and
E.
Still correct?
Yes, if you want, but I prefer to say No: I do not pursue to have a FIB of
the current classic design. My goal is to fill a table from which by means
of a single indexed/offset element lookup the next hop can be retrieved.
Heiner
|