[Babel-users] atomic updates and ipv6 cache flushing issues

[Leen Besselink]

On Sun, May 05, 2013 at 11:25:53AM -0700, Dave Taht wrote:
> On Sun, May 5, 2013 at 5:57 AM, Leen Besselink <leen at consolejunkie.net> wrote:
> > On Sun, May 05, 2013 at 05:11:26AM -0700, Dave Taht wrote:
> >> Robert bradley has been working on making atomic route updates work
> >> and possibly better route cache flushing.
> >>
> >> His current branch merges just fine with babeld head, and has tested
> >> it on 3.5, 3.7 and 3.8, but I have not tested it thoroughly yet
> >> myself.
> >>
> >> get the patchset from: git://github.com/rb12345/babeld.git
> >>
> >>
> >
> > Hi,
> >
> > I've been watching from a far what Babel has been doing and I find it all very interresting.
> >
> > Also I have some test VMs I'm running to see how Babel works and seen this bahaviour as well.
> >
> > I hope to find the time to give this code a try.
> >
> > My questions are some what related:
> >
> > Now that Linux 3.8 has support for IPv6 ECMP is there any plan to use that again to make updates
> > more attomic or even support multipath directly. Because that is what I've been interrested in.
> >
> > If I remember correctly Dave Täht also mentioned he would like to see someone implement it.
> 
> It's not ECMP for ipv6 I'm looking to see solved for multiple routing
> protocols, but the multiple ipv6 prefixes/multiple exits source
> routing problem. Take the case where you have one network but two
> gateways.
> 

Ohh, I see.

> GW A                                  GW B
> 2001:db8:2::1/48                   2001:db8:1::1/48
>          |                                        |
>         --your-hosts-on-your-network--
> 

I actually have a setup similar to that at home, with 2 tunnnel providers
and 1 router which acts like 2 gateways, it sends 2 RA/prefixes.

But it is more like failover setup, when one tunnel is down, it stops sending
the RA for that prefix. I have it set to expire very quickly.

So I totally understand why you'd want to do that.

> Both gateways distribute both prefixes to all clients, so they have
> their choice of prefix to use for any given connection attempt.
> 
> Solving for a single default gw in a routing protocol in this case
> won't work, as packets sourced from 2001:db8:1:X won't go out
> 2001:db8:2:X and vice versa.
> 
> The homenet folk demonstrated this form of ipv6 multihoming working
> (using a version of OSPF and BIRD, I believe) at the last ietf. It was
> very cool (to me at least) to see openwrt used in all three of the
> bits and bytes demos.
> 
> http://www.ietf.org/blog/2013/04/bits-n-bytes-on-video/
> 

Thanks, I'll have a look at that.

> -- 
> Dave Täht
> 
> Fixing bufferbloat with cerowrt: http://www.teklibre.com/cerowrt/subscribe.html