Most of what you describe Sander sounds reasonable, and sounds aligned
with what is i the deployment documents.
The WG debated the EID allocation extensively. One could argue that
there is no need for it, that we could merely request that PI
allocations be granted for LISP EID usage individually. The WG felt
that if we could get all IPv6 LISP EID allocations from a single block
that was not used for anything else, that would simplify avoiding
lookups in the mapping system that were inevitably going to fail. Thus
the allocations request.
Yours,
Joel
On 11/16/2012 4:12 PM, Sander Steffann wrote:
Hi Joel,
Why does any operator have a reason to carr any routes other than their paying
customers? Because it provides connectivity for their customers.
If we get this block allocaed, then it results in 1 extra routing entry in the
global routing table to support LISP inter-working.
An entry that some of their customers may use, whether the operator carrying it
knows that or not.
In fact, it would take significant extra work for the operator to somehow block
this aggregate.
If LISP fails, this is a small cost to find out.
If LISP succeeds, this results in significant reduction in core tabl sizes for
everyone.
That still assumes the altruistic routing of the prefix. And I am afraid that
if the usage of this block gets a bad reputation that all LISP usage will share
that reputation. I really like LISP but I am still not convinced that it's a
good idea. If we can find a way to get the EID prefix routed in a reliable way
then I would love it!
I really care about LISP and I am afraid that: people see unreliable routing for
EID /16 => assume LISP is unreliable. That is why I am pushing so hard to get
this sorted out.
Hmmm. What about the following strategy:
- Start with the EID prefix being handed out like PI
- Either through the RIRs if they are willing to take the responsibility
- Or from a separate registry
- Some altruistic /16 PITRs might show up in the global BGP table
- A holders of a (assume) /48 from the EID prefix can arrange PITRs for their
own space
- And announce it as a separate route in the global BGP table (for now)
- Keep the routing and reliability under their own control
- If the experiment is a success we advise ISPs to:
- Install their own PITRs for the /16
- Filter out the /48s at their border
The filtering of the more-specifics once they have their own PITRs will make
sure that they use those PITRs and that they will use the most optimal path to
the locators as soon as possible. It will also keep their routing table
smaller. If enough big transit providers offer /16 PITRs to their customers
then the more-specifics might be filtered on a larger scale.
So, summary:
The ways to reach a PITR would be
a) Run your own PITR (big networks, LISP users)
b) Use one from your transit(s) (smaller networks that don't have their own)
c) Use an altruistic one as last resort
As long as (a) and (b) aren't a reality the LISP users who don't want to rely
on (c) can run/rent/etc a PITR for their own space. I think the routing
community would really appreciate it if all those more-specific routes would be
temporary until wide deployment of (a) and (b) make them unnecessary.
And if this doesn't become a success we have a bunch of /48 prefixes to the
separate PITRs in the BGP table. It won't be much, otherwise we would have
declared success. So the risk of messing the BGP table up is very limited. And
when can tell people: if you are bothered by those more-specifics in your
routing table you can always deploy your own PITRs and filter the
more-specifics at your border. That might keep everyone happy.
What do you think?
Thanks,
Sander
