Hi Fred,

Comments inline.

Thanks,
Praveen

On 2/5/14 8:44 PM, "Frederic Detienne (fdetienn)" <[email protected]>
wrote:

>Hi Praveen,
>
>I suppose it would be convenient for you to minimize DMVPN to a "routing
>solution" however draft-detienne covers more than that.
>
>The remote access client case is fully covered. Without routing protocol.
>This point was clarified sometime in November (I believe) and
>draft-detienne was updated accordingly.
>
>Now back to the discussion, I think I now do understand precisely how you
>see the negotiation going. It turns out we support all these methods
>already and we know the complexity of it.
>
>I would like to verify a few more points because it is not trivial at all.
>
>What I am exposing here is that draft-sathyanarayan will quickly grow out
>of control as it tries to encompass all possible negotiation methods.
>
>In practice, it means only a couple vendors will be able to implement it
>decently and will likely clash quite hard. All this at the expense of
>users who are lured by failed promises.
>
>More inline on this topic then I will slowly move on to the next points.
>
>Thanks,
>
>   Fred
>
>(sent from mobile)
>
>> On 06 Feb 2014, at 00:31, "Praveen Sathyanarayan"
>><[email protected]> wrote:
>> 
>> Hi Fred,
>> 
>> I see draft-detienne¹s view of discovery of tunnel-peer is a routing
>> issue. So DMVPN solution predominantly a routing solution. So,
>> draft-detienne may work good for routing vendors.
>
>> But it is not a solution for non-routing vendors. As an example, mobile
>> devices increasingly can do more features like FaceTime/Skype/Hangout,
>> Airdrop. To do these, expecting mobile devices to run BGP protocol, to
>> discover a direct tunnel is not a reasonable expectation. Or for that
>> matter expecting a firewall device to run routing protocol, is not a
>> reasonable expectation either.
>> 
>> More comments inline.
>> 
>> 
>> Thanks,
>> Praveen
>> 
>> 
>> 
>> On 2/5/14, 5:45 AM, "Frederic Detienne (fdetienn)" <[email protected]>
>> wrote:
>> 
>>> We can no longer call that a simple solution. There are pieces that
>>>meet
>>> some of the requirements but are in contradiction with others.
>> 
>>> Similarly, draft-sathyanarayan offers multiple implementation or
>>> deployment systems (policy based or tunnel based) which are not
>>> compatible at protocol level. It means implementations have to cover
>>>BOTH
>>> methods to guarantee interoperability.
>> 
>> [PRAVEEN] I disagree. How will a spoke running OSPF and other spoke
>> running BGP interop in draft-detienne?  Your answer was, (previously in
>> virtual conference) was that it does not matter. Because, as long as Hub
>> supports both routing protocols, there is no issue. The same thing
>>applies
>> here. If you have route based implementation on a SpokeA and Policy
>>based
>> implementation SpokeB, it can still open a direct IPSec tunnel without
>> causing any interoperability. And of course, there is no need of running
>> routing protocol between the spokes, even in our solution.
>
>The big difference is here:
>
>domain1 can deploy a routed based network only by picking a vendor
>focusing on that method only. In this case, the TSi TSr may be 0.0.0.0/0
>0.0.0.0/0
>
>Domain2 may deploy a pure policy based implementation.
>
>Now how does domain 2 accept domain1's proposals ? And vice versa ?


[PRAVEEN] Hmm… May be you should re-read our draft. What our draft says
that the suggester/hub will suggest what traffic-selctor should be used
for establish the tunnel. So each spokes knows exactly what Tsi and Tsr
being negotiated. 


>
>As you explain below, the route based method probably has to fall back to
>a policy type negotiation.
>
>If domain1 admin made that choice, one can assume there is a reason. Now
>we end up with a double negotiation method.
>
>Also how does spoke domain 1 initiate to hub domain 2 ? How does that hub
>know how to narrow down the TSi to ? It does not know that spoke yet...
>
>What now if domain 3 wants GRE/IPsec ? I remember an early email saying
>it can be negotiated too.

[PRAVEEN] Again in Tsi and Tsr payload carry that information shortcut
suggestion from Hub. If GRE is supported in a particular domain, it should
be able to switch to GRE/IPSec.

>
>Draft-sathyanarayan possibly allows any protocol to be negotiated but it
>focuses on the policy based method and does not mention at all how
>interop is achieved.

[PRAVEEN] Sure if needed we will update with the information that is
required. This is why we are having discussion isn’t it? ;-)

>
>
>> To take a practical example: one domain may initially be rule based, the
>> other tunnel based. What happens when those domains must now cooperate ?
>> 
>> 
>> Both issues above will prevent proper cross-domain interoperability. In
>> particular, a "policy-based" spoke will not be able to talk to a "tunnel
>> based" hub as per this draftŠ it would take a decision as to which is
>>the
>> fallback mode and a dual implementation on at least one of the devices.
>> 
>> 
>> [PRAVEEN] No issues here. As long as you support rfc5996, and you
>>exchange
>> TSi and Tsr payloads, you are ready to setup a tunnel and forward the
>> traffic. On route based VPN side, you get TSi and TSr payload that
>>defines
>> the SPD entries. These SPD entries are bound to your tunnel interface.
>> Route that exist before the shortcut tunnel, still points to tunnel
>> interface. As this tunnel-interface is now has dynamic SPD entries, it
>>now
>> knows how to forward all/selective traffic to the shortcut tunnel. And
>>of
>> course the Policy based domain side it is straight forward. Where do you
>> see the issue? 
>
>So I presume TSi TSr of the routed spoke will be 0.0.0.0/0 0.0.0.0/0 ?
>
>This then gets narrowed down by the policy spoke ?
>
>We would create a tunnel per peer to achieve that. Your comment is
>interesting however as it raises a complexity.
>
>On a policy spoke1, the initial policy is
>
>10.0.0.0/24 10.0.0.0/8 to hub
>
>Then spoke2 is discovered hosting 10.0.1.0/24
>
>Now the SPD says
>
>10.0.0.0/24 10.0.0.0/8 to hub
>10.0.0.0/24 10.0.1.0/24 to spoke2
>
>I know implementations that will not behave well in this case. The SPDB
>does not guarantee a best match or sorted search.


[PRAVEEN] This is described in Section 5. It explains with example. Please
take a look.

>
>What should they do ?
>
>Thanks,
>
>  Fred
>
>> Thanks,
>> Praveen
>> 
>> 
>
>

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

Reply via email to