6PE (was: Re: Virtual hosting provider Linode announces v6 support)

Jared Mauch jared at puck.nether.net
Thu May 5 03:30:07 CEST 2011

On May 4, 2011, at 6:05 PM, Daniel Roesen wrote:

> On Wed, May 04, 2011 at 08:50:02AM -0400, Jared Mauch wrote:
>> We are all overlay networks, be it over some IEEE, ITU or other
>> transport, and ultimately an overlay network over the layer-1
>> fibers in the ground.  Each time you create an opaque tunnel you
>> make it harder to diagnose what is wrong.  Trying to figure out
>> who is dropping your GRE or IPIP or other traffic can be quite
>> hard.
> How do you do fancy traffic engineering and employ things like MPLS fast
> reroute etc. without 6PE?

I think some of the 'fancy' stuff you speak of we do but in our own way for v4.  Regarding IPv6, we have not had to perform any TE on this traffic as it's significantly less than the IPv4 traffic we deal with so has not been an issue.

I hate to speculate on the ratio, but despite that, I think that it's something like 5000:1 4:6 being a fairly made-up statistic based on flawed data in my brain.

Some co-workers and I found ourselves looking at a graph recently thinking that someone was ddos'ed as it was flatlined at the 10 level.  It took us awhile to realize that the scale was in Mb/s vs Gb/s.  How the years of thinking have changed.

I'm hoping to get collect some detailed numbers/ratios of IPv4:IPv6 traffic for the 24-48 hours before, and after (as well as during) the world IPv6 day for comparison.

> Yes, I have a strong dislike of 6PE too, especially due to the "AFI
> break" in 6PE BGP NEXT_HOPs (mapped addresses). I'm nagging vendors
> since... uhhh... many years to get proper MPLS control plane, but the
> standard response is the usual "no customer demand".

I think the big push will be when people realize that you're stuck with the 'antique' IPv4 for your router-id/everything selection and we've all moved to something else.  While most of these 32-bit numbers don't actually matter that much, the protocols that require them and make them configurable as an dotted quad will find people confused how to use them and make them work.  One day some new startup will only do ipv6 and will tell the vendor they won't accept the hardware without the ability to configure it as an ipv6 address.  Then it will come.  I don't think it's likely to happen earlier based on the pushback I heard about Wes Georges effort at IETF to point this out.  Many people still IPv4-only think, it will take time...

> And we're also still lacking knobs to disable TTL propagation onto MPLS
> labels just for 6PE traffic...

I've only seen one case where hiding hops had some significant value.  I seem to recall that in the win95/98 days it had a default ttl of 32 and when one network hid hops it made some 'faraway' locations no longer too far away to talk to some major web portals.

- Jared

More information about the ipv6-ops mailing list