How many IPv6 prefixes should you have (Was: IPv6)

Jeroen Massar jeroen at
Fri Nov 19 06:36:46 CST 2010

Job Snijders wrote:
> They are missing roughly 1000 prefixes. 


which just now when I peeked stated at the top:
2704 good/required prefixes
Minimum of 1714 prefixes (-990)
Average of 3513 prefixes (+809)
Maximum of 3861 prefixes (+1157)
186 peers, 67 not connected

As such, a proper transit should be delivering you with 2704 prefixes
for it to make you able to reach every place that should be reachable.

Some organisations send upwards of 1157 *MORE* prefixes than necessary,
note that is almost 30% more, thus junk prefixes which you don't need,
guess where a lot of those junk prefixes come from...

GRH doesn't have data on Cogent unfortunately (maybe check RIS?).

L(3) at the moment misses 632 prefixes that they should have:

For everyone's 'but another "transit" has more prefixes excuse, they
don't get all the prefixes either, they are missing 178 of them:

Though I have to note, for both the latter two cases that quite a few of
those prefixes are marked orange and thus those prefixes are only seen
by a small majority of the folks who send prefixes to GRH anyway.

What now is more disturbing is that there appears to be a couple of
prefixes out there which are not in the ARIN registry anymore which are
still being used (Hexago/Gogo6/Freenet6/nameoftheday's 2001:5c0::/32 is
an exemplary one) but also 2001:1890::/32 for AT&T worldservices,
2001:4800::/32 which once was Rackspace and quite a few others...

It just shows that everybody is just letting IPv6 routing grow as a
swamp till the real transits move in and start filtering nicely to get
the weed sorted out.

Btw RPSL anyone? You do know that nowadays the RIPE IRR allows ARIN
prefixes to be properly authenticated and registered too I hope?

See for more information:

Please use this facility, kthx!


More information about the NANOG mailing list