Dear Akamai, you got a /32 there not a bunch of /48s - how to break Facebook and annoy lots of users
Patrick W. Gilmore
patrick at ianai.net
Tue Aug 21 08:13:36 CEST 2012
On Aug 20, 2012, at 17:24 , Tore Anderson <tore.anderson at redpill-linpro.com> wrote:
> * Patrick W. Gilmore
>> We prefer to use a block from the ISP hosting the node
> So you prefer getting individual assignments for each of your nodes.
> Okay, with you so far.
Assignments implies too much. We just need addresses. We prefer they are SWIP'ed to us, but it is not a requirement.
>> When necessary, [...], we have to make a decision whether the node is
>> worth using our own IP space.
> Hmm. So if you prefer getting individual assignments in the first place,
> why don't get individual PI assignments as a second best option in this
Because we felt getting a /32 from each RIR and splitting as we please was quicker, easier, and cleaner. Plus it is completely within the rules.
Why isn't that a second best option?
>> i.e. when the hosting ISP cannot give us space
> Say *what*? Do these really exist - [IPv6 enabled] ISPs that can not
> provide their customers with [IPv6] address space? That sounds like
> something out of a thedailywtf.com post to me.
Never underestimate the power of human stupidity.
>> [...] come up with a better solution than asking for 1000+ /32s.
> So you haven't just come across *one* such absurd ISP, but over a
> thousand? That's good news for IPv6 deployment, I suppose - here I
> thought that I would be hard pressed to find over a thousand
> IPv6-enabled ISPs full stop - but these weird ISPs of yours must surely
> be outnumbered greatly by the normal ones that happily assign you
> address space...right? Perhaps you could just take your business to
> those instead? I've got space for you, racks AND addresses, if you need
> it... ;-)
First, 1000 blocks != 1000 ISPs. We frequently have multiple nodes in the same ISP.
Second, I was extrapolating from v4 numbers. It is entirely possible - in fact, quite likely - that was in error. The number is still non-zero, both because some networks are silly and because we have some multi-homed nodes which require our own IP space.
> Seriously though, you *can* go to the RIPE NCC and say in one single
> request «I've got 1000+ sites, please give me a /48 for each of them». I
> can't see any reason why such a request would be rejected. You'd
> probably get a nice contiguous /38 (shorter if you document a growth
> expectation) from the PI range, from which people that filter strictly
> allow /48s. Win-win.
Perhaps we should consider it.
I still don't think we've done anything wrong (other than mess up a few route6 objects).
More information about the ipv6-ops