Wednesday, November 19, 2014

Dual ISP, no BGP and IPv6 - what do my design options look like?

One of the recent questions I am getting around IPv6 deployment has to do with connecting remote sites and making branch offices highly available without using BGP. Today, with IPv4 it is not uncommon for a typical branch office to have two ISP connections, each with unique IPv4 ISP address space (provider assigned). The branch site will have a router or firewall that will make use of both connections by leveraging NAT and some sort of tracking or performance algorithm to determine which connection to use. It is able to do this because the solutions utilized NAT to obscure the actual IPv4 addresses used by the hosts. It is cost effective and works for many scenarios where some sort of higher availability and/or failover is desired for these sites without the complication of BGP. All these configurations allow the remote site to have local Internet hop off without backhauling the traffic across WAN or VPN links back to the corporate office.
So the natural question that comes from this IPv4 topology solution is, how do I do this in IPv6?
The answer is not as clean and neat as many would like it to be but there are several options you can utilize. Let's cover each of them and go over their merits and shortcomings.

1. Use both IPv6 provider assigned address ranges in your remote branch location. Optionally use NPTv6 on the router(s) to handle failure scenarios of ISP's and to provider faster failover.

2. Use a single IPv6 provider assigned address range in your remote branch location. Use NPTv6 on the router(s) to handle failure scenarios of that primary ISP. Simplifies host address configuration.

3. Use VPN to extend your primary provider independent address space to the remote branch for routing. Optionally use NPTv6 on the router(s) to handle local Internet hop off. Simplifies host address configuration.

4. Use ULA in your branch location and use NPTv6 on the router(s) to direct traffic to the appropriate ISP. Simplifies host address configuration however complicates routing and global reachability.

5. Upgrade the site to use provider independent configuration and do BGP and run a single prefix that is routable. No NPTv6, no PA address space and highly available.

After describing these options I am often asked, which is the best? What configuration should be used? The first option almost everyone selects is 4 - it looks and feels like what they operate with IPv4 today. I can't blame anyone for picking that initially, especially given it familiarity. But I believe ULA should only be used in narrow situations with an eye on preferring global unicast address space at all times. After picking the option of global unicast address space the next sorting preference is provider independent address space. This is a tough one and for many is not a viable option due to a variety of reasons. Costs, operational skills to run BGP and slightly more complex routing configurations, hardware requirements, plus often it can be political. So, with a preference for global unicast, likely provider assigned address space where are we left? Options 1, 2 and 3 are all still possible. My next rule is to try and simplify host address configuration and try and make RFC 6724 as predictable as possible. This means preferring a single prefix from one service provider to address the site. Just like in IPv4 you will likely prefer a particular connection due to speed, cost or some other criteria that is dictated by business decisions. Use the address prefix that comes out of that decision and address the site. Use a VPN connection from the other ISP to connect the remote site back to your corporate via routing and set up NPTv6 if needed. This solution address almost all use cases and has the advantage of being predictable from a routing and topology basis.
If you have existing PI space from a regional registry like ARIN then consider extending the address space to the remote site via VPN. If you require local Internet hop off you can then enable NPTv6 and leverage that. Remember, NPTv6 brings all the penalties of potential protocol brokenness. There can be situations where certain applications and protocols will not function correctly so NPTv6 is an undesirable tool for that reason. This is one of the reasons that I avoid ULA and why making use of ULA is so far down my list. At least with a true global unicast address you can guarantee proper IPv6 functionality in all cases.
To make things even more challenging - there are very few NPTv6 (or older NAT66) implementations available. So those that think NAT and IPv6 is a happy future, think again.
I hope this helps clarify how I come to my design recommendations and why I think ULA is a poor option for many situations and how leveraging NPTv6 today will be a challenge.
- Ed

1 comment:

Philip said...

I've considered the same problem and there are other problems introduced if you enable router discovery. In my case, I have two different routers connected to the two ISPs and so each host needs to know which router to use as its default route.

Nice if there was some neat solution.