[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: IPv6 support in d-i



Hi,

On Mon, Sep 24, 2012 at 07:08:47PM +0200, Philipp Kern wrote:
> > I have seen DS Light on Power-Point from big ISPs like DTAG and 
> > O2/Telefonica and i have seen it on PowerPoint from Vendors like Juniper,
> > Cisco and Ericsson/Redback. Yet i fail to find a good explanation what
> > DS Lite solves? What is the benefit to Dual Stack?
> 
> You tunnel IPv4 over IPv6. Hence the CPE only needs to communicate by IPv6 on
> the WAN side. The CPE also sees the internal IPs behind the CPE (no NAT44 on
> the CPE) and sets up a mapping from [IPv4, Port] to [IPv6, internal IPv4,
> internal Port]. So you save the double NAT44. Some CGNs seem to be able to do
> uPNP for port forwards, too (as the CGN sees the internal address in the CPE's
> LAN, it can forward the packets to the right place).

I'd be interested to see a solution for >= 64K subscribers supporting
uPNP for the subscriber side. Havent seen anything on Vendor slides.
Even ALGs (Application Level Gateways) for FTP, SIP, etc are too costly
to support so what i have seen is a VERY limited support.

But still - This is all not a single reason for DS Light. CGN can be done
with a PPP Dual Stack session aswell. And BTW - I dont think that the
NAT in the CPE will be eliminated. If you do, you'll have another bunch
of problems like prefix delegation for IPv4 or eliminate PPP and use
proxy arp e.g. bridging.

Another issue with eliminating the NAT in the CPE is that suddenly you
CGN address space shrinks very fast - You have a /10 - When you hand out
a /24 per subscriber you'll be able to support 2^14 subscribers per CGN 
instance. 16K subscribers is not much - Todays linecards support
16-64K Subscribers so you'll need up to 4 CGN instances per linecard,
and the big operations dont to CGN on the Edge but use a centralized
approach. I hear hundrets of MPLS VPNs creeping up to shovel Data from
the Edge to the CGN gateway. 
 
> But then obviously this is stateful. Cisco and the IETF seems to develop
> something called MAP which is stateless and maps v6 address info into v4
> address and port number. Crazy stuff.
> 
> > One reason this might get interesting is Vendor Licensing or Linecard
> > Codespace - A Dualstack session requires twice the amount of resources
> > on the linecards so all vendors halved their max-subscriber count with
> > Dualstack. If one shifts aways the CGN/DS Lite concentrator from the
> > BRAS one could save resources. But its just a matter of shifting
> > resources and money around. RFC1925 (6) It is easier to move a problem
> > around than it is to solve it. 1)
> 
> With DS-lite you only need one IPv6 session on the BRAS.

And where do you hold the IPv4 State? At least for AAA purposes?

As far as i understood the reasons for DS Light is licensing costs 
in the mobile backhaul as those are per PDN context e.g. backhaul channel
to the mobile. Those are today not capable to do dual stack (Although
Standardised in 3GPP Version 8/9) so you'd need 2 PDN contexts, one for
IPv4 and one for IPv6. So Mobile Operators thought of DS Light so
they'll not need another context and simply switch UAs to v6 transport
only. 

IMHO DSLight will be a niche market. With 3GPP Version9 we'll see Dual
Stack in the Mobiles and my guess is that we'll see v4/v6 dualstack,
later v4CGN/v6 Dualstack on DSL and FTTx. This will most likely be
necessary for something like 15-20 Years. My guess is also that
v6 will bring us the return of PPP as the access l2 protocol for
a lot of operators as making the l2 backhaup v6 safe e.g. ND
filtering/isolation is much harder.

Flo
-- 
Florian Lohoff                                                 f@zz.de

Attachment: signature.asc
Description: Digital signature


Reply to: