mkorourke at gmail
Jun 22, 2012, 8:56 PM
Post #10 of 11
Re: IPv6 BGP TE (was Couldflare routing problems)
[In reply to]
Sure, if you apply that definition that's another way to skin the cat; Cost
as a factor comes to mind on the below - obviously much cheaper to have a
single /32, yet being where we are and the lowest common denominator
tending to rule, obviously that just needs to be understood as a cost on
doing business, and as an architectural consideration for enterprises,
content providers or other.
One use-case that is valid globally for v6 TE and one I'm sure we're all or
will be impacted by - DDoS.
- Using TE to pull an impacted /48 or other /36 or /xy whatever is very
- Getting that Chinese, Russian, USA or international traffic away from
your other customers to be cleaned
- It's a use-case commonly used globally for v4 DDoS TE ie. advert of a /24
or multiples of to pull impacted traffic into a cleaning centre
- Yes off-ramping and advert of a /32 and de-avert in other places will
work - assuming you have the capacity, it's just not a desirable scenario.
- Perhaps it's a moot issue and the TE use-case will simply work we'll
enough with the majority of the v6 world accepting /48's;
On Fri, Jun 22, 2012 at 4:15 PM, Sascha Lenz <slz [at] baycix> wrote:
> > >> I specifically said I don't see a valid use case for *globally*
> > > advertising
> > >> more specifics.
> > â€¢ Single provider or enterprise allocated a /32
> > â€¢ Multiple independent global locations eg. PNG, India, New
> Zealand, US
> > â€¢ Each site advertising out a /36
> > â€¢ Requirement for each site is too large for a /48 allocation.
> > â€¢ Advert of a covering may have some very undesirable results
> > â€¢ Most networks it will work today will they'll accept the routes
> in the above approach.
> > â€¢ The approach has some parallels IPv4. People are going to do
> this and take v4 practice and apply to v6 if within general community
> guidelines - and why not, why re-invent the wheel?
> > â€¢ Networks likes yours be they teir2/3/4 that don't accept a full
> table in the v4 world would have a covering route with ACLs potentially
> used for bogons. Why not for v6?
> > â€¢ What's your suggestion as to BCP for the scenario?
> i'm not sure if i understand your points in the first place, but why would
> ISP have three separate networks but only use one allocation, and why
> would and enterprise work with a PA allocation instead of PI assignments?
> A single provider, if you mean an ISP has a single network, otherwise it's
> multiple ISPs by (my) definition, perhaps multiple local ISPs
> sharing the same name but not being the same legal entity or AS number.
> Each separate local/regional entity then would become member of the
> appropriate RIR and receive their own /32 or bigger allocation to
> aggregate the local traffic.
> A single enterprise is an end-user and doesn't aggregate assignments, so
> they can always apply for a separate PI assignment of the appropriate size
> from each responsible RIR, /48 or shorter.
> One of the differences between IPv4 is that de-aggregating an IPv6 /32 or
> shorter into /48 is much worse than even de-aggregating even an IPv4 /16
> into /24s - be it fat fingers or intentional Traffic Engineering shit.
> But perhaps you can enlighten me where i misunderstood your scenario.
> Mit freundlichen GrÃ¼ÃŸen / Kind Regards
> Sascha Lenz [SLZ-RIPE]
> Senior System- & Network Architect