[6bone] Is minimum allocation /64 now?

Pekka Savola pekkas@netcore.fi
Fri, 24 Oct 2003 09:12:48 +0300 (EEST)


On Fri, 24 Oct 2003, Dan Reeder wrote:
> Having read that rfc, howcome you suggest /112 or /64 rather than a /126 to
> get around the anycast problem? The section 4.3 clearly states that the /126
> will work fine - what is the point of suggesting a shorter prefix? To me
> thats just wasteful addressing.

Uhh, please stop to think about it.  Even if we use /112's, we can have
2^48 of them, assigned from a single /64.  No ISP should need ever more
point-to-point addresses than that :-).

Remember that "wasteful addressing" has entirely different meanings in
IPv6 than IPv4.  Once you have an IPv6 /64, you can put as many nodes in
that as you want, compared to e.g. an IPv4 /24.

/112 is a great simplification over /126 from the user's perfective.  This 
is because with /126 you should use something like:

3ffe:ffff:ffff::f00:{1,2}/126
3ffe:ffff:ffff::f00:{4,5}/126
3ffe:ffff:ffff::f00:{7,8}/126
3ffe:ffff:ffff::f00:{a,b}/126
3ffe:ffff:ffff::f00:{d,e}/126
3ffe:ffff:ffff::f01:{1,2}/126
....

We just dedided that we want to end the address with either "1" or "2"  
(we also have a methodology to determine which end of the link is given
which number), /112 gives the last 16 bits to a subnet, so this is
possible, like:

3ffe:ffff:ffff::f00:{1,2}/112
3ffe:ffff:ffff::f01:{1,2}/112
...

If this model was used towards the customers, /112 would add more 
flexibilty for future changes (e.g., the customer adds a firewall, /126 
can given an additional address which is mostly fine).

Seems simpler to me, and there's plenty of address to play with.  We
assign all point-to-point addresses from a single /64.

> Also, could you please clarify when a linux system would be deemed to be a
> router rather than a host? Or perhaps I misinterpreted the rfc and it only
> applies to routers such as ciscos?

It applies to all the nodes which act as a router.  This happens with
Linux, for example, if you have toggled on net.ipv6.conf.all.forwarding
sysctl (or done something that accomplishes that, like set up
IPV6FORWARDING=yes).

> We use /127s for the "point to point" tunnels and as far as I'm aware i've
> not seen any problems (other than redhat 9 always applying PREFIX::0/128 in
> the routing table to its loopback for some reason)

That's exactly the reason why /127 are not to be used between the routers!  
Between a router and a host, it _should_ be OK as long as the router is 
given the PREFIX::0/127 address.  But who can say when the other end will 
not be connecting a router or not?  Hence, /127 should not be used.

-- 
Pekka Savola                 "You each name yourselves king, yet the
Netcore Oy                    kingdom bleeds."
Systems. Networks. Security. -- George R.R. Martin: A Clash of Kings