[ppml] IPv6 flawed?

David Conrad drc at virtualized.org
Mon Sep 10 21:12:09 EDT 2007


On Sep 6, 2007, at 10:30 AM, Iljitsch van Beijnum wrote:
>> According to <http://www.conference.sanog.org/slides/conference/
>> sanog10-pfs-deaggregation-report.pdf>, we're seeing the
>> "deaggregation factor" increase "slowly and steadily since 'records
>> began'", with the fastest growth occurring in the "new" Internet (the
>> graph on page 15 is very interesting).
> Note that this is "prefixes with maximum achievable aggregation" vs  
> "currently seen prefixes". If you count the number of prefixes per  
> AS, you'll be getting the same number for a good number of years: 8.

Well, yes.  And if you truncate the deaggregation factor, you get 1  
since data collection started.  Looking at the historic data from the  
published routing analysis of prefixes per AS, you'll see a fairly  
rapid decent from 12.13 in Feb 1999 to 7.97 in Jan 2004 after which  
it has been increasing "slowly and steadily" ever since (we're at  
8.81 now).

Interestingly, if you look at the ratio of ASes that are announcing a  
single prefix to all ASes, it has consistently gone up (from 27% in  
1999 to 42% today).

I suspect both of these are causes for worry.  The first is likely  
driven primarily by people breaking up aggregates for good or bad  
reasons.  The second is likely due to multihoming.  Fortunately, the  
growth isn't that great right now.  The big question is how these  
trends will change in the future.

>> Since IPv6 uses the same
>> routing and traffic engineering technology as IPv4, I am curious what
>> constraints could be put in place to keep PI space down to about 1
>> per ASN.
> Prefixes per AS aren't limited by routing technology (if only it  
> could...) so this is irrelevant.

Yet you go ahead and address this irrelevancy:

> in IPv6, you'll be either getting a /32 or a /48 as a PI block. So  
> unless you manage to get multiple PI blocks, any efforts to inject  
> more than a single prefix will easily be thwarted by prefix length  
> filters.

Of course, this assumes people will implement and maintain prefix  
length filters.  I'm told by some in the ISP business that the  
economic pressure to remove such filters is sufficiently high that  
they don't believe prefix length filters are viable in the long term  
(I'm paraphrasing a bit).


More information about the ARIN-PPML mailing list