On Fri, Jun 22, 2012 at 02:42:24PM +1000, Rod Whitworth wrote:
| On Thu, 21 Jun 2012 18:52:18 -0500, Mark Felder wrote:
|
| >On Thu, 21 Jun 2012 18:39:24 -0500, Rod Whitworth <glis...@witworx.com>
| >wrote:
| >
| >> It is not a "school of thought" - it is how it is. I have seen one /126
| >> out in the wild but it is very lonely.
| >
| >I work at an ISP/datacenter. We use /126s for the link net. Handing out
| >/64's "because you can" is stupid in my worthless opinion :-)
| >
|
| It's not because you can, it's because it's best practice, it makes
| renumbering easier and most of all when you use /64s your subnet
| addresses are so easily readable.
"It makes renumbering easier" is a very poor argument. Renumbering is
just as easy wether you use /64s or /126s. Simply replace the first
64 bits and .. tadaa.wav .. you've renumbered.
It's not best practice at all. It's common practice. Doesn't make it
best. The fact that (older) RFCs told you not to do it is irrelevant,
there are now also RFCs that want to prohibit NAT for IPv6 - I'm not
sure what is more ridiculous.
| What do you have?
| /24 ?
| /32 ?
| /48 ?
| /56 ?
| All of the above have xx00:0:0:0:0:0 as the last part of the address
| and when you slice off /64s they all have 0:0:0:0 as the last four
| words so documenting is easy for any of your subnets.
You can also say: "This /64 is for point-to-point links" and then
document each and everyone in there by the remaining 64 bits. Further
class 'em up into customer-id (16, 32 or 48 bits) and line-id (48, 32
or 16 bits). Or split up even further. Either way will result in a
pretty sparse usage of subnets for point-to-point connections.
Or, use a /64 per customer, if that makes sense for you. Do what
makes sense, not what you read on the internet (unless the two match,
of course, which is often (but not always) the case).
| But I guess that being ultra-frugal with sunbnet prefixlen is really
| important for operators who have more clients than there are grains of
| sand on the face of the earth.
| That's roughly a /57's worth.
This remains a weird argument at best. If you get a /32 from your RIR
and every subnet *MUST BE* a /64, you can have "only" 4B subnets. Now
that seems like a lot, but what if you want to have some sensible
numbering in there, identifying customers, identifying VLANs,
identifying whatever. How many bits will you use for that ? What
makes sense ? Can you guarantee it's always going to fit in every
situation ?
You can have sensible, easy to understand and to explain numbering
schemes with v6. I have my doubts this is true for every environment
if you strictly adhere to the /64-per-configured-interface rule.
Oh, earlier in this thread you also made the link-local argument.
Funny. Note how that is *always the same* /64. There's deeper
reasons for link-local being a /64 than "because a network is a /64".
(Note that I'm not saying these reasons are good, Claudio!)
Paul 'WEiRD' de Weerd
--
>++++++++[<++++++++++>-]<+++++++.>+++[<------>-]<.>+++[<+
+++++++++++>-]<.>++[<------------>-]<+.--------------.[-]
http://www.weirdnet.nl/
