Thanks for the answers.

On 2/23/10 3:53 AM, "Ole Troan" <[email protected]> wrote:

> Reinaldo,
> 
> [...]
> 
>> Thanks for the answer. There seems to be some differences here that you can
>> help clarify in the context of anycast.
>> 
>> These questions are based on the following from the draft (and section
>> around it) that in my reading advances an idea of fully stateless
>> deployment.
>> 
>> "As 6rd is stateless, any BR may be used at any time. "
>> 
>> In 6rd although one can use anycast, there is a need for the IGP to be
>> stable, correct? It cannot be fully stateless otherwise different packets
>> for the same (source, destination, protocol) tuple would use different BRs.
>> Traceroute packets with the same TTL, source and destination taking
>> different paths would be a subset of a more general problem.
> 
> the IGP does not need to be stable. just like in a 'native' IP network you
> prefer that packets for the same flow follow the same path. when that doesn't
> happen, e.g because of link failure or IGP instability then the behaviour of
> 6rd packets will be just like the native IPv4 packets. the same thing happens
> if you withdraw the anycast route on BR failure.
> 
>> But in general what are the anycast affinity requirements for such a
>> deployable solution? But more specifically when connection based protocols
>> like TCP that react to delay are used on top of it.
> 
> there are of course operational issues here. and I would certainly not
> recommend load-balancing between BR's which are on different continents for
> example.
> 
>> I'm familiar with the use of anycasting for DNS where request and responses
>> are idempotent, short-lived and over a connectionless protocol, therefore
>> not much an issue.
>> 
>> And also use of TCP anycast as in this presentation
>> http://www.nanog.org/meetings/nanog37/presentations/matt.levine.pdf where
>> there are discussions on use and stability requirements.
> 
> right, but that's for using an anycast address as a endpoint for a TCP
> session. here we only use anycast as a next-hop.
> 
> this is not Internet-wide anycast like e.g 6to4 uses. this is all intended to
> be contained within an SP. where you also have tight control over the anycast
> advertisements in the IGP. not that I don't think this would also work in an
> Internet context, but then we're back to 6to4 with all its issues.
> 
> cheers,
> Ole
> 

_______________________________________________
Softwires mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/softwires

Reply via email to