Re: [v6ops] Operational Consensus on deployment

[Ross Chandler <[e-mail]>]

On 11 Aug 2014, at 08:29, Tore Anderson <[e-mail]> wrote:

> * Fred Baker (fred)
> 
>> On Aug 9, 2014, at 12:07 PM, Ross Chandler <[e-mail]> wrote:
>> 
>>> How about the case of an IPv6-only DC network with multiple
>>> separate SIIT-DC translation prefixes (mentioned in Sec 3.4). Which
>>> translation prefix will be used in the case where DNS64 (Sec 3.3)
>>> is also being used? It seems like DNS64 might only work easily with
>>> a single translation prefix.
>> 
>> RFC 6147 is designed for one translation prefix. One could imagine a
>> translation prefix per tenant, and you would (at least in concept)
>> need a DNS64 per tenant. That is unless you got smart and looked at
>> the source address of the packet to see which tenant it came from.
> 
> I'm thinking that multiple instances is an advanced topic that would be
> better documented in a separate draft. I'd preferably like to keep this
> draft concise and to the point, thus making it easy to read and understand.
> 
> In any case, multiple SIIT instances should be relatively
> straightforward. If the applications in questions are server-style ones
> that are merely responding to client traffic (think HTTP) then it's
> really a no-brainer, as the inbound client traffic will already have a
> source IPv6 address within the appropriate SIIT instance's translation
> prefix, to which the server will respond accordingly automatically.
> 
> If on the other hand it's the IPv6 application that is responsible for
> initiating traffic to IPv4 destinations, then it would need to choose
> the appropriate translation prefix for synthesising the IPv4-mapped IPv6
> address to initiate communication to. It would have to do that even if
> there is only one SIIT instance, too.)
> 
> If the application is using DNS, then DNS64 could be used for this
> purpose, which could be deployed in a number of ways (locally on the
> server itself, dedicated DNS64 servers per SIIT instance, one DNS64
> server but with multiple views per SIIT instance, ...).
> 
> Other options would be for the application to synthesise the IPv4-mapped
> addresses itself according to RFC6052, in which case the choice of
> translation prefix (and thus the SIIT instance) would be a configuration
> options; or finally, you could use the host agent approach, in which
> case the translation prefix (and thus the SIIT instance) becomes part of
> the host agent’s configuration instead.

That approach to documenting it sounds fine to me. 

Ross