The IP over ATM Mailing List Archive by date

[hiroshi@ctr.columbia.edu (Hiroshi Esaki)]

   >   cv > For IP multicast over ATM, you forward between LIS using routers.
   >  ohta>>  No, it is multicast but not multicast over ATM, unless the routers
   >  ohta>>  relay cells, that is, unless the conventional model is used.
   > 
   > No, I do not think so, 
   > multicast is the multicast of IP level, not of ATM level. 
   > So that, even routers of LIS relay multicast packets, it is fine for IP
   > multicast over ATM. 
  oht> With the same logic, you can say, even routers of LIS relay unicast
  oht> packets, it is fine for IP unicast over ATM.
Yes, I think so. 

  oht> Of course, it's fine if routers relay cells.
No, even when router does NOT relay cells, it is fine. 
The issue of such a usual router is its pakcet transmission performance. 


  oht> The problem is that, classical model is unfriendly to a lot of existing and
  oht> future Internet protocols and that, with cell-by-cell relaying routers,
  oht> we don't need the classical model.
As you well-know, networking with cell-relaying routers does not require 
any change from exsiting and future Internet protocol (e.g. RSVP). 
It is true, since it keeps catenet model. 
   But, I think, LIS will be adopted, whatever LIS may require some modification 
for the existing and future Internet protocol.  But, at least, since classical 
model keeps subnet model, it is much better than NHRP model, I think. 
Also, classical IP model can use celll-relaying routers for their LIS's 
routers, in order to improve the pakcet forwarding performance. 
And, of course, you can scale if LISes are reasonable size. 


  oht> So, the classical model is not wrong but is undesirable and unnecessary.
I can not say unnecessary, sorry. 


   > Both classical model and conventional model can relay the packets 
   > of multicast packet through cell-by-cell.   
   > It is just the implementation decision of each routers. 
  oht> But, if a router CAN NOT relay data cell-by-cell, it is NOT a router
  oht> for IP over ATM.
I do not think so.  ATM is just data-link technology. 
The similar thing is happen for frame-relay ot TDM-based networks 
internetworing.  These (e.g. frame-relay) neteworks can interconnecetd 
by routers, that genarally does not have data-link cut-thru function. 
Even in this case, it is router. 


   > When we only focus on the scalable IP multicast over ATM, classical 
   > IP model can scale, IF the LIS is not huge scale and the ATM cloud 
   > (the domain LIS can be defined) is not huge. 
  oht> I'm afraid you don't understand what we are discussing.
I am sorry when my argument is out of focus.....

  oht> LISes are, from the begininng, considered to be unscalable.
"I" think LIS itself is not scalable, personally. But, this may not be 
the general consensus, I think. 
Especially, Grenville says large scale LIS will be possible. 
I think well managed-size (number of LIS clients, number of LIS in the 
datalink segment, and geographical scale) LISes can scale, as well as 
the exsiting scaling-up by IP (e.g.through address aggregation). 


  oht> That is, everyone agreed from the day one that the number of hosts
  oht> in a LIS can not be arbitrary large.
OK, is this agreeable ? 


  oht> The problem we are discussing these days are that, LISes can't be
  oht> geographically large. The size of LISes are physically constrained.
  oht> The point of LIS model that the size of LISes can be arbitrary large,
  oht> is denied.
I think, LISes should not be geographically large. 
But, when the number of these geographycally large LIS is small, 
it will be acceptable. 


Regards, 

Hiroshi Esaki