On Wed, Jun 15, 2016 at 8:55 AM, Linda Dunbar <[email protected]> wrote:
>
> NVO3 Participants,
>
>
>
> I2NSF (Interface to Network Security function) has a work item in defining 
> the flow security policy between domains (which includes inquiry of the 
> capability from one domain to another and the actual flow policy rules from 
> one domain to another).
>
> Very often, the paths (or links) among nodes of a overlay network are 
> provided by other network operators (a.k.a. “underlay network”).  The flow 
> policy rules are intended to filter out unwanted traffic from underlay 
> network so that various attack traffic won’t saturated the access links to 
> the overlay nodes.
>
>
>
> One interesting scenario brought up is Overlay nodes may need to request some 
> traffic to be traversing IPsec channel. To achieve this goal, it is necessary 
> for Overlay Network controller to inquire if the needed IPsec resource are 
> even available before send the request (may even involve AAA process between 
> controllers of each corresponding domain ).
>
>
>
> Want to have a survey if people see the use case of Overlay Network needing 
> portion of traffic to be through IPSec channel?

Yes, this is a valid use case, and one that we  are looking at as well.

> IPSec is supposed to be between two end nodes. Here we assume that the 
> Overlay nodes don’t have the resource or capability for IPsec, but expect 
> IPsec between flow’s ingress and egress nodes (i.e. NVE).
> Any opinion is appreciated.


>
> Are there any use cases of overlay network needing IPSec among their nodes 
> only for a specific time span? i.e. Time based IPSec connection?

Time based IPsec connection is not a use-case we have encountered.
People usually use IKE for periodic key-rollover, if that is the goal.

However, applying IPsec to specific flows (e.g., those defined by a
src or dst port on which the service listens) is important.

But that also made me wonder about the interaction between IPsec/IKE
and the proposed BGP FS (IPsec is frequently used between end-systems
that do not want to run a BGP daemon). Since the config information that
needs to be distributed are things like keys, algorithms etc to populate the
sadb/spd, IKE looks more appropriate in most cases.

Like [CJ], I too have to read the draft in greater detail to comment further.

--Sowmini

_______________________________________________
nvo3 mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/nvo3

Reply via email to