In einer eMail vom 10.11.2009 16:26:22 Westeuropäische Normalzeit schreibt  
[email protected]:

> From: [email protected]

> Yes, but do  hierarchies right.
> ...
> I have my  doubts that the routing folks have a proper understanding wrt
> hierarchies.

This is probably true, because we just haven't had  that much experience 
with
them; the _very_ limited amount of hierarchy in  use in the current system 
is
just not enough to fully educate us about  them.

(I'd like to make an analogy to congestion control - back in the  '77-'79 
time
period, not only did we not know much, we didn't know how much  we didn't
know... Well, a few people probably had a clue, but the rest of  us did not
really... :-)

But I do think this particular topic is  something that the RRG _should_ be
thinking about, since it's closer to  what I think a 'routing research' 
group
should be thinking  of.


Fine. I can promise: It is not just about reducing some ugly aggregation  
material. 
It is also about enabling Moore's law. And even better: Given that you can  
index the best next hop information by ONE offset info as taken from the  
packet header (similar to MPLS), you can also do temporary reroute by the 
same  speed: by just storing the temporary alternate next hop info at that 
place where  as a default the very best next hop info would be placed. (plus a 
long list of  more advantages).



(The  location/identity separation stuff is not, to me, really routing -  
it's
more basic architecture, but since the I* doesn't seem to have a forum  for
basic architecture discussion, it seems to have wound up  here...)


I fully agree.   


> What it takes is a "sliding hierarchy" as  provided by TARA so that each
> router is fairly in the  middle of the hierarchy and never at the very
>  rim.

Sorry, I don't know enough about TARA to respond to this. Your  Istanbul
analogy didn't really help me much. Can you say  more?

Well, being a godfather of PNNI you ought to understand the  
Istanbul-analogy :-)
 


> Hierarchy is by no means a reason to enforce  a path which is longer
> than the shortest  one!

True, but.... it seems to me fairly fundamental that the basic  idea behind
hierarchy and routing is that it allows one to discard _some_  information, 
to
make the amount of data one has to process in the  path-selection 
computation
more reasonable. That being the case, it seems  inevitable that _some_ 
stretch
(over the theoretical minimum) should  result.
While being in the USA you should really discard the info about  streets 
inside European villages (better: you should not even get them). At  the same 
time you should definitely NOT discard any physical link which directly  
interconnects some place in America and another place in Europe. In the end, it 
 is better to speak of a hierarchically sparsed flat topology !
So I use the term hierarchy only because  this term is better  understood 
by the readers (imho)

Noel wrote:
To return to your real-world map analogy, nobody planning a trip from  
Madrid
to Gdansk goes out and obtains a map that shows _every_ road in  Europe, and
does their path-selection based on that. Any realistic approach  would use
'high-level maps' from which many of the minor roads have been  omitted.
However, ignoring those minor roads may result in a path which is  longer
than the theoretical minimum.
 
This depends on your concept. If you want to go from Istanbul-West to  
Istanbul-East and your hierarchically upper map recommends you to go to Ankara  
(or even Teheran) first before being able to appreach  Istanbul-East then 
you will have this ugly stretch factor.


Noel wrote: (And yes, I know, a theoretical-minimum length path would  
probably be slower
than a path which took the larger limited-access  highways... most people 
want
to optimize time, and the road system itself has  been designed to optimize
that goal, e.g. with the constuction of those  larger limited-access 
highways,
so in the real-world there's an interesting  interaction between the
map-abstraction process and the infrastructure  planning process - maybe the
network world needs to do something  similar?)
 
Well this is state of the art. To play with the weight values of the  edges.

Heiner

_______________________________________________
rrg mailing list
[email protected]
http://www.irtf.org/mailman/listinfo/rrg

Reply via email to