[arin-discuss] Trying to Understand IPV6

Owen DeLong owen at delong.com
Mon Sep 13 21:27:20 EDT 2010


On Sep 13, 2010, at 5:59 PM, Alan Batie wrote:

> On 9/13/10 4:23 PM, Owen DeLong wrote:
> 
>> This is a really horrible approach... Yes, if you give a /48, it's 
>> gone. So what?
>> 
>> If you need more /48s, then, get another /32 or even a shorter than 
>> /32 prefix.
> 
> One of the apparent goals is to *avoid* going back to the bar for
> another block, though I believe the registries are doing the same sort
> of thing for the same reason: that way it will be *possible* to get a
> "shorter than /32"
> 
Correct... The RIRs are doing allocation by bisection.

>> For sparse allocation, allocate your /48s by bisection...
> 
> same idea, just moving the left edge and doing slightly more complicated
> math.
> 
Except it eliminates the whole right-bits/left-bits thinking and becomes
pretty straightforward and easy to automate. Also, since you aren't
doing variable-length allocations, you don't have to worry about
fragmentation.

> I should say digits rather than bits though, as it simplifies thinking
> about it:
> 

Well, this approach simplifies the human thinking until you start filling
things in, but, comes at the disadvantage that you hit higher densities
faster.

> 1xxx: customers
> 1000::
> 1100::
> 1200::
> 1300::
> ...
> 1010::
> 1020::
> ...
> 1110::
> 1120::

Owen




More information about the ARIN-discuss mailing list