CloudFlare IPv6 BGP announcements - WTF guys?

Daniel Roesen dr at
Tue Jul 17 03:57:29 CEST 2012

On Mon, Jul 16, 2012 at 09:56:00PM +0100, Brandon Butterworth wrote:
> On Mon Jul 16, 2012 at 09:15:17PM +0200, Daniel Roesen wrote:
> > Wether you see /32 PA more-specifics from all the CDN nodes, or PI /48s
> > doesn't make a difference at all, technically.
> It makes a huge differnce.
> With the latter the /48's will be from a range we expect /48's
> from (hopefully). With the former it's from a range we'd rather
> not have to leave open to /48's. Slightly classful can be useful.

But that's not a technical difference (impacting FIB scaling), but an
operational difference (to which I fully agree).

OF COURSE the "right thing" would been to use a separate PI block per
CDN site.

I fail to see a technical reason not to do that, but I see multiple
economical and practical reasons.

On one hand, get a /32 for a one-time effort of becoming LIR and a
yearly (for Akamai etc.) low fee in the low 4-digit range and number
up to 2^16 CDN nodes from that without ever having to discuss with
the RIR. Oh, you can actually design some hierarchy into that /32 as

The alternative: go to the RIR for every fscking new CDN node, submit
requests, wait for approvals, perhaps having to discuss nonsense with
newbie-IPRA-de-jure (time is money!), and pay 50 EUR [in RIPE case]
per anno per CDN node. Ah, and have all that tunnelled thru a LIR of
least mistrust which probably will also want to get paid for that
service. :)

Pretty convincing arguments to just shoot for getting and deagging
the /32, eh? Wait until the beancounters understand that concept. :)

Best regards,

CLUE-RIPE -- Jabber: dr at -- dr at IRCnet -- PGP: 0xA85C8AA0

More information about the ipv6-ops mailing list