[ppml] Fwd: Keeping in reserve

Andrew Dul andrew.dul at quark.net
Thu Oct 5 17:28:55 EDT 2006


>  -------Original Message-------
>  From: Kevin Loch <kloch at kl.net>
>  Subject: Re: [ppml] Fwd: Keeping in reserve
>  Sent: 05 Oct '06 13:11
>  
>  Iljitsch van Beijnum wrote:
>  > So what is the downside of simply giving out an additional, non-
>  > growable prefix when a /48 isn't enough?
>  
>  This results in involuntary fragmentation just like slow-start
>  in IPv4.
>  
>  > Or just give out the space that would have been reserved immediately,  
>  > i.e., give out /44s instead of /48s?
>  
>  That provides more space with which to voluntarily deaggregate.  That
>  could be a feature or a problem depending on your perspective.  It could
>  also be seen as unnecessarily wasteful since /48 is already a very
>  generous minimum assignment.
>  
>  I like the bisection method.  It provides the maximum ability to expand
>  delegations without fragmentation.  It also removes any perceived
>  relationship between reserved space and your delegation.

In general, bisection seems like a good strategy.  The problem with bisection from my perspective is what happens when the super-block you are assigning from becomes close to "full."  

In this case ARIN has decided to assign IPv6 PI from the block 2620:0000:/23 (http://www.arin.net/reference/ip_blocks.html).

Because of the IPv6 policy agreed to between IANA & the RIRs.  ARIN can't go back and get more IPv6 address space until they have assigned/allocated/reserved 50% of the space they have been given.  At least that is how I read the policy.  (http://www.icann.org/policies/proposed-ipv6-policy-14jul06.htm) 

If you are assigning /48s without reservations then you force bisection down to the /47 level.  If any org wants to grow bigger than a /47 then you force fragmentation by not having a reserve.  The IPv6 policy between the RIRs & IANA recognizes reserved address space as "unavailable" for the purposes for obtaining additional address resources from IANA.

The reserve method isn't perfect, but I believe it is reasonable based upon what we know now.

Andrew




More information about the ARIN-PPML mailing list