Hi Ole,

See inline.

Cheers,
Ian




>Ian,
>
>> Sorry to dig this one up again, but although there has been a fair bit
>>of discussion around the topic, I don't think that we have so far
>>reached an agreement on how to provision sw CPEs with the address of the
>>concentrator for MAP-E and lw4o6. We need to get a final answer on this
>>as there's a few drafts (softwire-unified-cpe, softwire-map-dhcp,
>>softwire-lw4over6) that are relying on it.
>> 
>> To try and summarise what's already out there:
>> The Unified CPE & lw4o6 drafts propose the use of DHCPv6 Option 64 (the
>>DS-Lite AFTR FQDN option)
>> Map-dhcp describes the use of the default mapping rule (DMR) to provide
>>the /128 of the BR
>> 
>> I think that it's reasonable to say that both work ­ I.e. They contain
>>enough information to provision the client with the /128 of the
>>concentrator so that it can send traffic.
>> 
>> Here are the arguments for/against each that I've seen. I've tried to
>>list all the ones that I've seen, so if I've missed any, then please add
>>them:
>> 
>> DHCPv6 Option 64
>> Pros
>> FQDN means that DNS round-robin load balancing could be used
>
>RFC6334 specifies that only the first AAAA should be used.

[ian] Checking 6334 section 5 - "If any DNS response contains more than
one IPv6 address, the B4
   system picks only one IPv6 address and uses it as a remote tunnel
   endpoint for the interface being configured in the current message
   exchange."

So you could use Option 64/DNS to provision multiple endpoints. You'd need
another mechanism (e.g. The bfd model) to detect failure and take
advantage of it, though.

>
>> Reuse of an existing option
>> 
>> Cons
>> Can not be extended with additional sub-options
>
>the CE needs to deal with DNS TTL and the added complexity of having the
>resulting "lifetime" of the
>default router be different than the lifetime of the address/mechanism.
>
>only one BR can be configured
>
>> map-dhcp DMR
>> Pros
>> Option has an 'encapsulated options' field, so could be extended with
>>additional functionality (although no additional functionality has
>>currently been proposed at this stage)
>> 
>> Cons
>> Only one DMR /128 can be provisioned to a client, so round robin load
>>balancing can't be used
>
>you can use anycast.

[ian] You can. But then, why do you need the list of addresses suggested
below? 

>
>> May conflict with the MAP-T (</128 prefix) DMR ­ Could be considered to
>>be out of scope for this discussion, but if we are following the
>>mechanisms described in Unified CPE so that a CPE can work out which SW
>>mode to configure, then a dedicated MAP-T sub-option (not shared by any
>>other mode) is needed. This is related to conversations that I've had
>>around extending the Unified CPE model for other softwire approaches.
>> 
>> My view: If there is additional functionality required in this option,
>>over an above the provisioning of a /128 prefix for the concentrator,
>>then the MAP DMR (limited to MAP-E) is the clear choice. However, if no
>>such functionality is needed, then we should re-use Option 64.
>
>does anyone know why a FQDN (with restrictions) was chosen for DS-lite?
>
>considerations:
> - MAP: should BR be inside or outside of the domain. if inside use an
>IPv4 address, if outside use an IPv6 address.
>   (I believe we have agreed on outside, and for LW46 it must always be
>outside).

[ian] I think it's simpler just to use outside as a single approach.

> - should the CE have multiple BRs? and if so, what should it do between
>them? load-balance?
>   the choice in 6rd was to allow for multiple BRs, but not specify how
>multiple BRs could be used
>
>my preference is for a simple list of IPv6 BR addresses. i.e. none of the
>two above options.

[ian] As long as this is done as a separate option to the DMR, then I
think this would work.

>
>cheers,
>Ole

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

Reply via email to