On Wed, Nov 12, 2008 at 5:31 PM, Scott Brim <[EMAIL PROTECTED]> wrote:
> On 11/11/08 6:29 PM, William Herrin allegedly wrote:
>> On Tue, Nov 11, 2008 at 4:38 PM, Scott Brim <[EMAIL PROTECTED]> wrote:
>>> On 11/11/08 3:39 PM, William Herrin allegedly wrote:
>>>> http://bill.herrin.us/network/rrgarchitectures.html
>>> Since you start by talking about the root cause of the routing scaling
>>> problem, I'll start by saying that it isn't _just_ conflation of locator
>>> and identifier.  Even if identifiers and locators were completely
>>> separate, if routing stayed the way it is today you would still have
>>> problems because sites will still want PI prefixes if only to avoid site
>>> renumbering problems, and will still inject prefixes for traffic
>>> "engineering" and to prevent hijacking.
>>
>> Thanks Scott.
>>
>> Isn't the renumbering hassle tied to the fact that the addresses
>> identify the servers?
>
> As far as renumbering goes, external access to servers can get sorted
> out through DNS TTLs,

Hi Scott,

Operational experience says no lifeline with DNS TTLs I'm afraid. DNS
TTLs only work right with applications which pay attention to DNS
TTLs. This works out to something south of 10% of deployed apps: most
interface with DNS via the gethostbyname call. Some undetermined and
indefinite time later, the other 90% are restarted or encounter some
other event that causes them to look up the DNS entry again.


> but I'm told there are also problems because
> changes are required in internal routing and forwarding configuration,
> and in configuration of external sites you have special relationships
> with.

I'm not sure that makes sense in context. Given a protocol in which
the GUID, SID and LOC are separate, why would you ever statically
assign the LOC to anything? Borrowing the MAC address example again, I
can count on the fingers of one hand the situations in which they get
statically assigned, especially across multiple routing domains.


>> Would you expand on translation? How does translation reduce the
>> number of entries in the routing table?
>
> Consider Six/One Router or NAT66 or LISP's "NAT".  If you translate a PI
> prefix into one or more provider-aggregatable prefixes, you've
> eliminated the PI prefix from global routing.

If I understand Six/One router (and please correct me if I'm wrong),
it alters the the IPv6 source and destination addresses entering the
core by replacing the edge PI prefixes with provider-assigned
prefixes. The edge PI prefixes are fed (via the mapping system) to the
router where the packet exits the core. When exiting the core at the
other side, the edge PI prefixes are restored.

I'm not sure I'd call that translation. It seems like Strategy A
variant A1c where the "provider-assigned prefixes" are RLOCs. Instead
of adding space in the packet for the RLOC, a portion of the GUID is
mapped out to make room for the RLOC on entering the core and then
mapped back in on exiting the core. The extra map-in-on-exit is the
price paid for maintaining compatibility with the existing IPv6
protocol.

Maybe that's an engineering compatibility difference like the
gyrations for PMTU in map-encap rather than an architectural
difference? What do you think?

How do NAT66 and LISP's NAT differ from Six/One Router?

Regards,
Bill Herrin



-- 
William D. Herrin ................ [EMAIL PROTECTED]  [EMAIL PROTECTED]
3005 Crane Dr. ...................... Web: <http://bill.herrin.us/>
Falls Church, VA 22042-3004
_______________________________________________
rrg mailing list
[email protected]
https://www.irtf.org/mailman/listinfo/rrg

Reply via email to