Rafa, Gabriel, and Sowmini, 

Here are some comments from IPSecMe group for your 
draft-abad-i2nsf-sdn-ipsec-flow-protection-01. I think it worth the check



-----Original Message-----
From: Yoav Nir [mailto:[email protected]] 
Sent: Thursday, April 13, 2017 4:42 PM
To: Linda Dunbar <[email protected]>
Cc: [email protected]; Michael Richardson <[email protected]>
Subject: Re: [IPsec] Can IPSec (RFC 5996) support tunnels with end point being 
(virtual) CPEs which has a set of workload attached (say Virtual Machines) all 
having virtual IP addresses?

<<snip>>

>> - I2NSF has a proposal on using Controller to manager all the IPSec
>> tunnels:
>> https://datatracker.ietf.org/doc/html/draft-abad-i2nsf-sdn-ipsec-flow-protection.
>> What kind of issues do you see with the proposed approach?
> 
> I didn't read it.

I did. They have two cases. In case #1 the controller provisions credentials 
for IKE and entries in PAD and SPD. In case #2 you forego IKE and have the 
controller provision the SAs (including keys).

I especially didn’t like case #2. Sharing a secret key among three entities is 
a bad idea. A shared authentication credential can also be misused, but that’s 
a hard attack to mount. A shared traffic key makes the controller a very 
attractive target.

More in general, SDN was born in the data center. In a data center an 
all-knowing controller makes sense. This is true for routing as it is for NSFs 
such as firewalls, IDPs and IDSs. VPN extends the reach of the private network 
to all corners of the Internet. Think of a store chain with a CPE in every one 
of thousands of branches. Or a bank. The problem there is that there is no 
central administrative function. Local branches may switch ISP and renumber 
their network without bothering to tell the IT people. So the model where the 
controller knows everything is tough to deploy in practice.

It is probably necessary to have two-way communications, where the CPE tells 
the controller about its topology (how it partitions the Internet to “in” vs 
“out”) so that the controller can set up the appropriate SPDs.

There have been several attempts at this. RFC 7018 describes requirements, but 
the WG ultimately failed to publish a solution document.  There are also more 
recent commercial solutions sold today under the marketing name of “SD-WAN”, 
which is sort of like SDN if you squint hard enough. All of these have some 
interaction between CPE and controllers (or hubs) which draft-abad does not.

Yoav

_______________________________________________
I2nsf mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/i2nsf

Reply via email to