INET6 required for SCTP in 7.0?

[Andy Dills]

On Thu, 6 Mar 2008, Mark Andrews wrote:

> 	Service providers get paid to push IP packets.  They shouldn't
> 	care which protocol version is in the header.  What they
> 	should be worried about is ensuring that they are here in
> 	4 years time.

Sure they should. The ASICs in the vast majority of production routers are 
setup for IPv4. Add in the fact that you can get very capable routers 
reasonably cheap on the secondary market and compound it with the lack of 
revenue driven demand, and economics overwhelms.

Very precisely because we are worried about being here in four years time, 
we spend our money wisely. We spend today's money today. Throwing money at 
something with no demonstrable or projectable ROI is exactly how you wind 
up gone in four years.

> 	Most end users won't even know that they are running IPv6
> 	connections.  I had to look at netstat to see which protocol
> 	was being choosen on my father's box.  I'm sure he had zero
> 	knowledge that he was using IPv6 (6-to-4).

This is true, but illustrates my point. If users had to be dragged kicking 
and screaming into using digital television, which is obviously a huge 
upgrade that provides a significantly enhanced experience, why would they 
want to pay for a new CPE that works fine and will work fine for many 
years? Which also in turn provides them with more IP addresses than they 
can use via NAT? 

> > - To route IPv6 with the same features and packet forwarding rate as with 
> > IPv4, nearly every network will be forced to purchase expensive router 
> > upgrades with no other real benefit beyond IPv6 connectivity (which again 
> > provides no ROI to justify the capex). Nobody is going to do forklift 
> > upgrades just for IPv6, but as routers get normally upgraded IPv6 
> > functionality will indeed slowly expand.
> 
> 	And the same arguement was put out 6 years ago.  The backbone
> 	really has gone dual stack while you wern't paying attention.

Portions of it, yes. But this is expected; "the backbone" frequently has 
to upgrade for a variety of reasons, ranging from new and valuable 
technology (MPLS, DWDM, etc) to shady behavior by Cisco (forcing people to 
get the SUP720-3BXL to handle >255k prefixes).

Every step you take away from public corporations who are spending 
stockholder money and have revenue driven infrastructure upgrades, you 
move toward companies who have a much slower growth rate with much fewer 
changes in network requirements, and who have to get capex approved by the 
person who's money is actually being spent on the improvements.

> > - IPv6 provides almost no technological upgrades beyond additional address
> > space. DHCP addressed the auto configuration feature, VPNs addressed
> > IPsec.
> 
> 	That extra address space really is a big advantage.  It
> 	really is so much better to be able to get to machines you
> 	need to without have to manually setup application relays
> 	because you couldn't get enough address space to be able
> 	to globally address everything want to.

So much better? Sure. Does it justify IPv6? I'm not convinced. 

I'm hoping some genius devises a new protocol that solves the growing 
issue of inter-domain routing scalability by eliminating the need for 
forwarding paths for every prefix in the global routing table, while also 
creating true network portability, allowing individuals to obtain personal 
IP space which they can utilize independant of their service provider, 
without requiring any knowledge of routing protocol.

THAT is worth a forklift upgrade. THAT would be rapidly adopted. 

IPv6 at this point looks very poorly thought out in the face of such 
obviously incremental solutions such as:

- Utilizing the rarely used 16 bit Identification field or the useless 32 
bit Options field in the existing IPv4 header to include a private routing 
identifier.
- Existing routers are compatibile, as they merely route the /32 to the 
NAT device, don't care about those fields.
- The NAT device rewrites the packet based on the private routing 
identifier, without user intervention in configuring mapped addresses or 
ports.
- The private routing identifier can either be a new DNS record or stuffed 
into TXT records.

Initially, "important" devices would not rely on the private routing 
identifier, enabling fringe users to use as a "best effort" upgrade while 
network stacks and resolver libraries get upgraded. All software upgrades, 
all leaving the core untouched.

That's just something I threw together while responding. Imagine what 
could happen if somebody smart focused on it.

> 	So make the network IPv6 enabled.  Both my home network and
> 	the office networks have bee IPv6 enabled for years now.
> 	My ISP doesn't support IPv6 yet though I know that have
> 	IPv6 netbocks for themselves now if not for the customers
> 	at this stage.

Oh, they have them for the customers. They just don't want to upgrade 
their routers.

> 	There is a reasonable chance that this mail will leave here
> 	over IPv6 for some of the recipients.  It will almost
> 	certainly travel over IPv6 for at least one hop.

s/IPv6/uucp/ ;)

Andy

---
Andy Dills
Xecunet, Inc.
www.xecu.net
301-682-9972
---