Re: Granularity (was Re: [RRG] ALT + NERD is inelegant & inefficient, compared to APT or Ivip)

[Brian Dickson <briand@ca.afilias.info>]

David Conrad wrote:
> Brian,
>
> On Jan 23, 2008, at 4:11 PM, Brian Dickson wrote:
> > It may be that rather than having one problem/solution match-up to 
> > consider, that there are two:
> >
> > dual-homed
> > multi-homed (N>2)
> >
> > I'd argue that the requirements of each differ, and there can be 
> > significant scaling benefits from splitting them out, and handling 
> > each separately.
> ...
> > What do folks think?
>
> My gut feeling (FWIW) is that this would be a mistake.  The 
> assumptions that you make differentiating the N=2 vs. N>2 cases appear 
> to be subjective and dependent on life as we know it now, not life as 
> how we might reasonably project it to be. 
The assumptions were derived from two premises:
(1) the end "site" is a consumer-grade router, designed to operate with 
nearly zero knowledge
(2) the EIDs need to operate analogous to ULA. No conflicts, but no need 
to manage them, assign them, or otherwise add process or management to 
handling the need for EIDs

Whether the resulting design meets the needs for which it was designed, 
is the criteria that should be used for evaluating it.

None of the design aspects were built on any significant assumptions 
about life as we know it now.
> What is to say that in the 'near' (for some value of that variable) 
> future, I won't want my cellphone router connecting my PAN multi-homed 
> to the (say) 6 cellular providers it can find signal for?
I'd say that:

   * you are not the average user :-)
   * you can certainly use the N>2 method without any problem
   * nothing is stopping you from doing the heavier-weight stuff
   * it might be feasible to implement with multiple EIDs and
     source-address selection on the host

I wouldn't expect this to be used for any "mobile" devices, only for 
fixed (e.g. hard wired, or fixed wireless) sites, which I am presuming 
is going to be the usage driver for the bulk of 2-homed sites/hosts.

The presumption is that not a significant number of homes who are 
willing to shell out bucks for redundancy (or have their employers do 
so), would pay for a third link, if two gets them redundancy.

And add to that that in many, perhaps most, markets, there are likely to 
be two last-mile providers of layer-1 access (telco and cableco), 
regardless of who does the layer-3 stuff (DSL ISP using telco facilities).
> I'm not sure that coming up with an architecture that builds in these 
> sorts of assumptions won't end up biting us painfully in the end.
Perhaps. But using a one-size-fits-all approach has resulted in missed 
opportunities.

What's going to be a challenge in addressing the need for host 
granularity or near-host granularity (home site) in LISP are the following:

   * big ISPs (cable and phone companies) may not want their customers
     to have this, as it lets their customers have control, and compare
     provider performance
   * whatever is done, the software/data footprint will need to be
     small, since consumer grade devices aren't likely to be
     upgradeable or have huge amounts of RAM
   * both demand and uptake will be dependent on "consumers"
   * uptake will be gated by device manufacturers for the consumer market

And most importantly - for LISP to work, it needs ITRs everywhere 
(between hosts and ETR-connected hosts).
It works best if ITRs are close to or at the site which is multihomed, 
which means consumer-grade ITRs.

Brian Dickson

--
to unsubscribe send a message to rrg-request@psg.com with the
word 'unsubscribe' in a single line as the message text body.
archive: <http://psg.com/lists/rrg/> & ftp://psg.com/pub/lists/rrg