Search RPD Archives
Limit search to: Subject & Body Subject Author
Sort by:

[policy-wg] IPV6 PI space

Bill Woodcock woody at pch.net
Thu Dec 14 19:52:32 UTC 2006


      On Thu, 14 Dec 2006, Andrew Alston wrote:
    > I firmly believe that with regards P.I space, a /48 allocation is enough.
    > Firstly, for IXP space, there is a trend to move towards /126 networks on
    > point to point allocations (do not use /127's, there are a variety of
    > reasons for this), therefore, IXP's could operate off a single /48,
    > utilizing /64's per IXP, and utilizing /126 networks for the actual peering
    > requirements.

While the peering would be directly across the /48 or /64, rather than 
across a bunch of point-to-points (which, if needed, would be supplied by 
the peers, rather than the IXP, if done following the practices used in 
the rest of the world), I can verify Andrew's assertion that a /64 is 
what an IXP needs.  A /48 is, therefore, also sufficient.  And for the 
sake of simplicity, I'd strongly recommend just making all v6 PI 
allocations be /48s, period.

    > Secondly, with regards to end user networks, for AfriNIC to allocate /32's
    > in P.I space would again run contrary to what the global trends are, ARIN
    > has agreed to /48 P.I space, and (someone correct me if I 'm wrong here),
    > APNIC is about to ratify their /48 P.I policy.  

Yes, that's correct.

    > These need a small, non-wasteful, low maintenance allocation - for 
    > example, an allocation of a /44 would allow for 16 "networks" (each of 
    > /48) to be established.

This is incorrect.  An IXP, by definition, is a single subnet, across 
which all participants can reach each other.  Therefore, it need never be 
more than a single /64.  A /44 is wasteful, since only 1/2^20 of it, or 
0.00009% could ever be assigned.  And bear in mind that even then, the 
largest exchange points the world has ever known have had fewer than 500 
participants, so in terms of actual IP addresses used, as opposed to 
subnets assigned, at _best_, that would be 0.00000000000000000000005% 
utilization.  Again, not very efficient.  I don't think you need to be 
worrying about giving IXPs room to grow.  They have plenty.

    > A /48 is more than suitable for a single (even geographically 
    > dispersed) IXP

There's no such thing, again by definition.

    > a /44 should be sufficient for a country or a countries ISPA.

Since no routing aggregation is possible, there's no benefit to assigning 
IXP subnets for a country in a nominally-aggregatable manner, and this is 
a false economy, resulting in additional wasted space.  And there's no 
inherent connection between IXPs and ISPAs; in fact, exactly the opposite.  
The connection in South Africa is a fluke which the market hasn't yet
corrected because there aren't sufficient market forces at work there.

So, briefly, what you could do, which is what's working elsewhere, is to 
stick to the basis of what Andrew is suggesting: make all PI allocations 
/48s, and leave it at that.  Nice and simple.

                                -Bill




More information about the RPD mailing list