Re: [v6ops] draft-ietf-v6ops-enterprise-incremental-ipv6 WGLC

On 12/08/2013 17:27, Arie Vayner (avayner) wrote:
> Owen,
> 
> While the arguments about moving the firewalls closer to the users are valid they are often are not practical (or at least the customers I worked with would not implement this option).
> Imagine an enterprise network with 300 spoke sites, but only 2 or 3 Internet gateway locations (with some private WAN in between).
> Moving the firewalls to the spoke sites would increase the number of firewalls from ~3 to ~300 (I am ignoring redundancy and scale for a second)... This is a major CAPEX and OPEX impact...

Clearly DOS and scanning protection has to be done as close to the Internet
border routers as possible, and there your logic applies.

However, as Steve Bellovin pointed out many years ago, the best number of
firewalls for upper layer protection is one per host, which scales nicely
and has less CAPEX and OPEX than middlebox firewalls will ever have.

Not that I see any of this argument as relevant to the IP version number.

   Brian

> Arie
> 
> From: Owen DeLong [mailto:owen@delong.com]
> Sent: Friday, August 9, 2013 12:17 PM
> To: Arie Vayner (avayner)
> Cc: Eric Vyncke (evyncke); Lorenzo Colitti; Fred Baker (fred); v6ops@ietf.org
> Subject: Re: [v6ops] draft-ietf-v6ops-enterprise-incremental-ipv6 WGLC
> 
> 
> On Aug 8, 2013, at 22:21 , Arie Vayner (avayner) <avayner@cisco.com<mailto:avayner@cisco.com>> wrote:
> 
> 
> Another loosely related point that I think could make sense in such a document would be the ways to accomplish multi-homing and how it is different than today's IPv4 implementations.
> 
> Many enterprises rely on NAT on the Internet edge as their multi-homing/traffic engineering mechanism with IPv4.
> 
> If we recommend against ULA+NPTv6 (or just NPTv6 for traffic engineering), then we need to highlight the symmetry requirement due to stateful security layers.
> Traffic leaving from an Internet gateway site to the Internet has to come back through the same site, or the stateful firewalls would break the flow (well, has to hit the same stateful security layer)
> 
> Or stateful firewalls have to get better about sharing state. There are two things that can help with this...
> 
> 1.         Put your firewalls as close to the end systems they protect as possible. Make your security zones relatively small and place the firewalls closer together at those narrower borders.
>             This will often require more firewall units, but it helps in a number of ways:
>             A.        Firewall policy tends to be much simpler (and as a result less error prone and more reliable)
>             B.        The hardware demands on the firewall tend to be lower so you can buy cheaper units.
>             C.        The simpler rulesets can be more easily tailored to meet business requirements as they evolve.
> 
> 
> 
> 2.         Improve firewalls. Give the firewalls that all protect the same boundary a way to mesh-peer with each other and exchange information about the state tables such that triangle routing is no longer problematic.
> 
> 
> Syncing upstream and downstream routing policies is not always an easy task (but could be relevant in some cases).
> Linking the Internet gateway layer across sites (before hitting the stateful security layer) could be another solution.
> 
> If we make the changes above to the firewalls, this could be  a lot less relevant in most cases.
> 
> 
> Do you think a short discussion to raise awareness for this potential issue could be relevant in such a document?
> 
> It's certainly worth documenting. I'm not sure whether it belongs in this document or not.
> 
> Owen
> 
> 
> 
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> v6ops mailing list
> v6ops@ietf.org
> https://www.ietf.org/mailman/listinfo/v6ops