[address-policy-wg] Policy proposal: #gamma IPv6 InitialAllocation Criteria

On 20-apr-2005, at 18:05, Cameron Gray (RIPE Address Policy WG) wrote:

> My, now somewhat aging, e-mail to lir-help asked what happens to  
> LIRs that cannot get/justify/plan for 200 /48s their reply was  
> simple: get it off another LIR.  This now leads into a problem with  
> routing policies: if /32s are only to be allowed in the backbone,  
> how doe these sub-lir allocations get announced.

Well, there are no rules as to what is and isn't "allowed in the  
backbone". There is RFC 2772, which provides "6bone backbone routing  
GUIDELINES" and there are is a remark in the RIR/IANA IPv6 policies  
that the RIRs will only allocate blocks of /32 and bigger for the  
benefit of prefix length filtering. (But some root DNS servers have / 
48 blocks...)

At the end of the day, it's individual ISPs who decide what they  
allow and don't allow in their routing tables. It's true that if you  
announce a smaller block than a /32 you'll be filtered in many  
places. However, this isn't necessarily a problem. For instance, if  
someone in Asia sends your customer who has a /48 from you and also  
announces this /48 to another ISP, and the Asian network filters the / 
48, the packet will flow towards Europe as per your /32. Then when it  
gets to Europe, it's pretty likely that the packet will hit an ISP  
who actually has the /48 in their routing table. After all, what's  
allowing a few /48s from the people you get drunk with at all those  
RIPE meetings?

If then the link between you and your customer is down the packets  
still get to the customer. (Obviously if you drop the /32  
announcement for whatever reason this is no longer true.)

So I would encourage you, your customers and your customer's other  
ISPs to announce these smaller blocks over regional exchange points.  
This will give your customers 80% of the advantages of having PI  
space with only 20% of the downsides.

(Note that it would be good if those customers had a bigger block  
than a /48. Even a /47 would be better, as it's likely that ISPs will  
leak /48s that shouldn't be announced at some point, which will make  
others want to filter /48s. But since customers generally don't have  
bigger blocks than a /48 there is no reason to disallow /47s in anti- 
leak filters.)