Re: VCC cost models

[Andrew Smith <[e-mail]>]

> From [e-mail] Thu May  2 06:45:02 1996
> Date: Thu, 2 May 1996 09:30:08 +0500
> From: [e-mail] (Joel Halpern)
> To: [e-mail], [e-mail]
> Subject: Re: VCC cost models .... (was Re: Limits on SVCCs)
> Cc: [e-mail]

Joel,

> The recent discussion has been suggesting that a cut-through would
> "cost more" than a routed path across an NBMA fabric.

One word, Joel: "SCALING".

> Maybe I am looking at this differently, but it seems to me definitional
> that a cut-through costs the same or less in real terms.

Yes, ONE short-cut path costs approx the same as ONE hop-by-hop path.

> In the non-cut-through case, the packet enters the NBMA, and is passed
> over the NBMA, stopping to be handled and processed by several routers
> along the way.  In the cut-through case, it enters the NBMA, is passed
> at a "lower" level of proocessing across the NBMA, and exits and the
> ssame place it would have.

Agreed, in theory (and we can argue about best/worst current practice until
the cows come home) it is cheaper to process ATM VPI/VCIs than IP prefixes, 
although "lower" isn't always "cheaper".

> One can argue about whether the switches or routers are cheaper, or
> whatever.  But the packet is still entering and exiting the NBMA at the
> same point.  There certainly may be charging differentials, but I do
> not see what it is a given that a long-distance low level VC would cost
> more than a routed path.  

The real issue is that the individual legs of the hop-by-hop path can be
shared by many flows whereas the VCCs between the switches can only be
shared by those flows originating/terminating in the same place. So when
you have many flows that are going roughly in the same general direction
there are a lot of cost savings by using a hop-by-hop routed path. If you
have either no locality or total locality (many flows going between exactly 
the same ATM endpoints) then short-cuts might gain the edge.

> There are many possible charging models for
> IP service, and many possible charging models for lower layer service.
> If the pair are inconsistent, then sometimes cut-throughs will be a
> bigger win than they should, and sometimes they will be a big loser.
> Most of the time, I would expect the cut-through to be a small winner
> in terms of real resource consumption, but that is open to argumentation.

So I guess we are back to the old status quo: NHRP (the protocol) can be
used to request a short-cut NBMA address; NHRP (the internet-draft/RFC)
does not specify VCC usage policies, when to do make such a request or what 
to do when you get the answer.

Whether ROLC should also work on these other issues is a good question: I guess
that was the gist of Derya's original question. I know that MPOA's scope
involves providing some answers for one particular network deployment scenario
but that is not the Internet.


> Yours,
> Joel M. Halpern				[e-mail]
> Newbridge Networks Inc.

Andrew

********************************************************************************
Andrew Smith					TEL:	[phone]
Bay Networks, Inc. 				FAX:	[phone]
Santa Clara, CA					E-m:	[e-mail]
********************************************************************************