In einer eMail vom 20.10.2010 18:35:01 Westeuropäische Sommerzeit schreibt  
[email protected]:

Hi  Heiner,

Thank you very much for commenting.

> 3.1. Improved  routing scalability
> It lacks naming the two aspects in the first  place: memory consumption 
and CPU-performance (update churn). Only thereafter  it makes sense to 
request independency from the internet user population. Why  is the control 
plane 
mentioned specifically? I don’t understand. It is neither  necessary nor 
correct.


The control plane subsumes the memory,  bandwidth and CPU required for 
routing, both within individual systems and  network-wide.  We distinguish from 
the data plane, which is the transit  traffic.  There are similar 
scalability issues in the data plane, but the  argument there is much more 
challenging 
to make and is not widely  accepted.


Ok.I thought RIB=Control-, FIB=Data-plane and that using the term  
control-plane wouldn’t get us any further. Nevertheless the issues should be  
mentioned and specified: The current enormous size of the Internet ( number of  
DFZ-routers =x, number of RIB-entries = y, number of FIB-entries=z). The  
continuing growth wrt size and tighter meshing(FIB-growth-factor=350,000 /  
310,000=1,13).The expiration of IPv4 addresses (number of unused addresses=  n)


>  3.2. Scalable support for traffic  engineering
> Bad English: The solutions should also be scalable also  wrt detouring 
routes. ( and not wrt to traffic engineering).


By  detour routes, I'm assuming that you're talking about FRR and similar  
technologies.  This would seem to be an independent issue and I don't  think 
we have any agreement that this is an inter-domain  issue.


But for multihoming there is agreement. And multihoming is  just a special 
case of alternate routing, i.e.FRR (if seen  logically; I know   according  
to IETF terminology these are completely different animals). My guess is: 
FRR is  enabled intra-domain, hence there is a demand for it, 
intra-domain-wide. FRR is  not enabled inter-domain, hence there is no demand 
for it, 
inter-domain  wide.
 
.


>  3.3. Scalable support for multi-homing
> It  wouldn’t be bad to mention that multi-homing is just a particular 
aspect of  multipath.


Sorry, I'm not seeing your point.  Does multi-path  support affect our 
design goals?  


See above


>  3.4. Scalable support for mobility
> We should also mention support of  PRIVACY (where is the roaming user at 
this moment). Privacy can always be  supported (i.e. by any solution) by 
employing a home agent who will or who  will not relay. We may need to deal 
with both options. The current text  fosters the belief that   some solution 
> may provide both  concurrently.


Agreed.  Added a comment:  

"Mechanisms that help to provide more
efficient  and scalable mobility support are desired, especially
when they can be coupled with security, _especially privacy_, and
support topological changes at a high-rate."


>  3.5. Simplified renumbering
> The best of all is not mentioned:  solutions which do not require any 
renumbering.


Fair enough.   Added.


> 3.6. Decoupling location and identification
> I  have strong doubts that this community has fully understood what 
loc/id-split  may mean.


This doesn't seem like a document where we should be  educating them.

Agreed:-)



> 3.7. First-class elements
> Half of this  paragraph is used to explain this weir headline.
> What is really  intended here, should be expressed simpler and without 
this  climbim.


Please propose text.  I realize that my background is  unique, but I know 
of no better way to accurately and concisely describe this  in polite and 
professional language.  Saying that "the architecture can't  be a kludge" or 
"no cheesy hacks" doesn't seem to be appropriate.   ;-)

I did read 3.7 a few more times. I assume it is much more general  i.e.not 
just about tunnels. Myself, I cannot even find another example for what  may 
be meant. Also, if I were asked to come up with a first example, I would  
never have pointed to tunnels



> 3.8. Routing quality
>  A lot of words in  order to get to the term “stretch”.
> I have my own suspicion why this  term had been invented (and why it has 
been adopted). I can imagine that  routing quality may also mean different 
aspects (QoS,…)


If you  would care to propose text, I would be happy to replace the current 
 definition.  This is taken from the academic literature, where they  
wanted a metric to evaluate routing quality.  Personally, I don't  consider it 
to 
be a very practical measure, as it ignores policy and the  granularity is 
poor.  A stretch of 2 is simply  disastrous.

 I cannot propose text immediately because it needs more discussion up 
front. Quality routing may
 I cannot propose text immediately because it needs more discussion up 
front. Quality routing may
address a) the applied algorithms, or b) the selection of first-class paths 
and links (QoS-paths/links).
Ad a)  (only) : It is up to the provisioned information and up to the 
algorithm itself how good or bad is 
the resulting routing technology.
1. Imho, it is NOT quality routing if the second best route is  –
eventually- completely out of scope.
2. My suspicion: “Stretch” was introduced by Dima while judging  his own 
hyperbolic concept in 
comparison with the status quo. Shortest path means shortest path, and any 
longer path should be 
as short as possible and as long as necessary. Going from A to B via C, 
rather than directly from A to B,
means stretch 2 (which wouldn’t be such disastrous:-).Thinking about 
concepts with some conceptually 
built-in stretch factor >1: 
(1)FEDERAL EXPRESS: All parcels are sent to Memphis,TE, then forwarded 
(2)PNNI with its logical group nodes’ topologies.
We may say: The concept itself should not enforce longer routes. Comments?  
3.We are far off from better routing which e.g. might enable state-less  
p2mp /mp2mp, or scoped broadcast.4.We are far off proper dealing with loops:  
crankback means loop. Eventually crankback provides the only available route 
!  Quality routing may mean “better routing”



>  3.9. Routing security
> No comment to this  place holder phrase.


Your sarcasm is not constructive.  If you  dislike the content, please 
propose alternate text.



The text is ok. I should have written “sentence” instead of “phrase”.



>  3.10. Deployability
> The first sentence can  be deleted (blabla only).
> Any solution which required a flag day is  out from the beginning.
> Yes, incremental deployability is definitely  required.
> However, I think that ALL solutions can provide this (all  the so far 
displayed as well as all not-yet born  solutions:-)


Clearly, this is not the case.  It is very clear  that one can propose a 
solution that there are non-deployable flag-day  solutions.  While they are 
not going to fall into anyone's definition of  common sense, it is incumbent 
on us to be clear about the requirements  anyway.  In fact, in the past, this 
was a significant point of  discussion.


Agreed.


>  Furthermore I miss 3.11: 
> The solution should also enable Moore's Law  as addressed in RFC4984


I'm sorry, I don't understand your  point.  As has been argued in 4984, 
Moore's Law is not the only  constraint that we must work with.  DRAM latency, 
for example, is NOT  subject to Moore's Law.  Care to propose text?
Will think about some text.



Regards,
Tony

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

Reply via email to