Re: [lisp] LISP EID Block Size

[Rene Bartsch <[e-mail]>]

Am 2013-10-31 22:36, schrieb Geoff Huston:
> On 31 Oct 2013, at 11:57 pm, Dino Farinacci <[e-mail]> 
> wrote:
> 
>> Geoff, LISP can route the entire allocated address space (but just not 
>> requires it to be everywhere). So arguably, LISP can do this at much 
>> cheaper cost and complexity.
>> 
>> The reason for a dedicated block is similar to why we have an address 
>> block for IPv4 and IPv6 multicast. To experiment to see if a 
>> hard-coded block can provide any interesting ideas.
> 
> 
> On the LISP page I already see LISP using IPv4 and IPv6 blocks for
> this experiment. So what have you found out already in terms of
> "interesting ideas"? What do you think that a larger block would
> inform you that is not possible with the existing block?
> 
> (using /56 end site prefixes a /32 can accommodate 16,777,216 end
> sites of course, and even at a 10% utilisation level thats 1.7 million
> end sites. So I;'m kinds mystified why a .32 can't tell you about
> scaling given that we are talking of the order of 10**6 end sites
> within such a block.)

The success of the internet is based on experiments which directly 
scaled into production (like the IP protocol by Cerf and Kahn). Remind 
the pain we have with the IPv4->IPv6-transition because there are not 
enough IPv4-addresses for the whole internet-community. If you just go 
for tiny experiments - like the last 6 years - LISP will never go on air 
for the internet-community. Currently AVM provides a great chance to go 
an air by integrating LISP into millions of consumer-routers 'til end of 
the year and we shouldn't scare them away by a never-ending experiment.

If consumers adopt LISP, we'll need 10,000,000,000 prefixes within the 
next ten years. That means each public PITR will have to announce 
10,000,000,000 BGP-routes to itself if consumers use random PI-prefixes 
from the global unicast space. Multiply the number of public PITRs 
necessary for such a deployment with 10,000,000,000 BGP-routes each and 
you'll realize LISP will either break the internet and go into decline 
or just be a limited luxury for big companies. Using a dedicated 
PI-prefix which can handle 10,000,000,000 subnets will reduce the 
BGP-routes a public PITR has to announce to ONE. So using 24 or 32 bits 
is way to less for the expected growth.

I want to ask everyone on the list: Which facts prevent a scaling 
experiment with the aim of global production state? In my opinion a 
/16-EID-prefix is perfect for that goal.

Best regards,

Renne