[arin-discuss] ipv6 technology supplier phone bank?

[Ted Mittelstaedt]

Lee Howard wrote:
> I'm not sure this is still an arin-discuss thread.  I'm happy to move it
> somewhere else, unless we quickly get back to an ARIN-member
> question, like, "How should ARIN spend its resources to help?"
> 
> 
>>> ISPs will break IPv4 for their paying customers, and charge to put it
>>> back
>>> together?    Brilliant!
>>> Especially if the guy down the street hasn't broken it yet.[1]
>> It is not my plan, it is what the ISPs have backed themselves into by
>> stalling their IPv6 deployments.
> 
> IPv6 in the core is easy(ish).

Yes, and I think that today the core hardware is ALL IPv6-capabable, 
even if the administrators running a specific router don't have IPv6 
configured on it.  Thus, getting increased IPv6 penetration is
just a matter of ARIN continuing to harp on it to those operators.

> IPv6 in the data center is easy(ish).

Yes, and I think that today MOST stuff running in the data center is
IPv6-capable, and what still isn't, is going to be only used by
devices inside a particular data center that are dual-stacked.

That is for example, you can have a webserver querying a SQL server
that is IPv4-only, while the web server serves up both IPv4 and IPv6
to the public Internet.

Essentially, the data center is ready for IPv6 today, even if the
administrators running gear in a particular data center don't have
IPv6 configured in it.  ARIN could help IPv6 penetration here by
maintaining an "IPv6-certified list" of networks that sell connectivity
to those datacenters.

> IPv6 at the edge is hard.  It's partly a network value problem: consumers
> won't pay more for gateways and gadgets that support IPv6, so vendors
> don't make them.  
> 

It's much more complex a problem than that.

Virtually ALL CPE devices made today are HARDWARE-CAPABLE of supporting
IPv6.

The problem is that the vendors don't write the firmware to be IPv6-capable.

This is really a critical point because it's only been very recently
that this has been true.

I would say most CPE gear manufactured in the early 2000's did not
have enough ram and flash to support IPv6 stacks.  They could barely
support IPv4 stacks, and a great amount of gear also had what I would
call severely underpowered CPUs as well.  Who can ever forget the
infamous early-version Linksys BEFSR41s which arguably were one
of the worst implementations of an IPv4 router in history?

Most CPE of course, are built on all-in-one ASIC designs to save
money during manufacture, where the CPU/RAM/FLASH/RADIO/PORTS/WHATEVER
is all integrated into a single chip.

The CPE gear originally made in the early part of the decade used
general-purpose parts.  Take apart for example a Cisco 678 sometime,
this is a DMT-DSL modem.  They are stuffed with chips.  Whereas
a Cisco 8xx router with an integrated DSL port in it, has a much
lower parts count, and the unit has vastly more ram/flash and CPU
power.

That's why the IP stack on a 678 is IPv4-only, and the stack on
a Cisco 8xx is IPv4/IPv6-capable.

Thus, EVEN IF a vendor WANTED to write IPv6-capable firmware for
CPE gear back in 2004, for example, THEY COULDN'T.

One of the very first CPE's that IPv6 firmware was available on
was the early model Linksys WRT54g, revisions 1 - 4.  Revision
1.0 came out in 2002.  These revisions had 4MB flash/ 16MB ram,
and early IPv6 firmware was available from Earthlink, 
(http://www.research.earthlink.net/ipv6/)

While it is true that one big reason this CPE was selected by
Earthlink for it's prototype IPv6 firmware was
because of Linksys's use of GPL code on it exposed the programming
interface, the other was that this CPE was one of the few of that
era in the "cheap" category with 4MB flash.  Later on when
Linksys went to 2MB flash on these unit revisions, it killed the
ability to run IPv6 on them - no 3rd party firmware that supports
IPv6 exists that will run in that small flash.  However, there are
much better/faster/more ram CPE's (like the WRT54GS) today that
have superseded this older CPE

Nowadays, for 3rd party IPv6 firmware, the Linksys prototype IPv6
firmware is mainly of historical interest, most people who are
doing 3rd party firmware for IPv6 are running openwrt, or dd-wrt
firmware.  A large number of CPE's will support this firmware
as well.

> 
> My point was more that CPE vendors don't sell to ISPs, they sell to 
> consumers, who really don't want to be bothered.  ISPs can tell what is
> needed (work in progress) but don't generally send POs.
> 

This isn't accurate.  When it comes to CPE gear, Broadband-providing 
ISP's generally fall into 1 of 2 categories (I'm going to ignore dialup 
customers):

1) Mammoth ones, like your comcasts, earthlinks, and the telco ones
like Qwest, Verizon, etc.

These ISPs exclusively use all-in-one CPE devices that they have
CPE vendors (actiontec, westell, motorola, etc.) build to spec,
that include the DSL/Cable/Fios port integrated into the CPE along with 
a router that can be configured to either route a subnet or do NAT
or be a dumb bridge.

They then hand these out to new subscribers on various "free promo"
plans where the customer gets the CPE for nothing as long as they
have service a particular length of time.  For example Qwest charges
the subscriber for the CPE, then spreads the payment over the first
several months of service, while simultaneously running a 1/2 price
discount on the service itself - so in essence the subscriber
pays full rate on the service and gets the CPE for free, under a
gimmick intended to prevent the customer from assuming the CPE
is really free or owned by Qwest.

If these ISPs wanted IPv6 they would tell the CPE vendors (ie: westell,
actiontec, etc.) that IPv6 is part of the RFP for the CPE device,
and the IPv6 code would get added at pretty much no additional
cost nowadays - because the cost of writing the firmware for the
devices is a very small cost of the raw materials for the device
itself.  Not to mention that the CPE vendors could easily take the
existing openwrt/ddwrt source code, all of which is under GPL,
and modify it, which would cost virtually nothing to them.

Thus, switching the edge over on THESE ISPs is very easy - it's
merely a matter of convincing them that the next RFPs they hand out to 
the CPE vendors must require IPv6.  Then, the natural churn will
insure that once the IPv6 CPE's are deployed, that over time the
subscriber base will become IPv6.

The same issues exist with large wisps, (like Clear wimax) it is
just an issue of them specing it in their RFP's for CPE gear.

2) medium and small ISPs deploying over DSL, either
telco-supplied ADSL networks, or their own SDSL networks over dry pairs. 
  Since they do not have the buying power they can't go to the CPE 
vendors and make them add IPv6 code.  What they do is buy CPE's that 
have "generic" firmware on them and then resell them to customers.
Quite often the CPEs are used, or they are integrated into another
device (voice cards, like the Cisco 827-4v) or they are bridge-only
(many SDSL modems are like this, ie; Paradyne, old copper mountain
networks stuff, etc.)  Essentially the only possible way they can
get their edge customers IPv6 is to only use DSL modems that can
go into bridging, and use additional ethernet-to-ethernet firewalls
behind these modems.

Since these guys are competing against the
large ISPs, there's tremendous price pressure on them already
for the CPE - and so it's highly unlikely they are going to
want to do that.  Thus, switching their edges to IPv6 is NOT
easy at all.

Small "community" wisps are in an interesting boat - those that build
using openwrt or dd-wrt compliant gear are in good shape, that
that did not are going to be even less interested in switching
over.


So really, the issue of edge CPE's is thusly, you
should recognize that a vast number of edge customers use the
ISP-supplied CPE gear and do NOT purchase additional routers that
they would use behind such gear - and with those customers the
CPE vendors are indeed selling to ISPs.  The remainder do use
CPE gear but they are buying it across the board, some are
getting cheap consumer gear, others are getting expensive
industrial gear (ie: Cisco PIX, or ASA, etc.)

> How should ARIN spend its resources to help?
> 

If ARIN really wanted to make a dent they would have to be very, very
aggressive.  If I was king of the world, in charge of all RIR's, I
would immediately cease issuing IP addressing to ALL large ISP networks 
that did not agree to MANDATE that their CPE vendors include IPv6 code
in their new CPE models within a year.

In essense, no more IP addressing of any kind to Verizon, Qwest, 
Comcast, SBC, yadda yadda yadda until those companies signed an
agreement stating that from now on, IPv6 will be a required component
of all new Cable/DSL/Fios/Wireless CPE gear that they supply to the
retail Internet access market.  Give them a year to have their
CPE vendors get their house in order and then let the boom drop
if they don't do it.

That will get the edge IPv6-compliant.  However, getting the customers
to USE IPv6 will be a different matter entirely and it is intimately
tied up with the fate of the Windows operating system.  Although Win XP
"supported" IPv6, it was NOT enabled by default - so IPv6 was a 
non-starter for the masses.

IPv6 -is- enabled in Vista, but there were some initial problems with
it and some network adapters, there's also the "localhost" issue with
it.  Unfortunately this spurred a number of "how to disable IPv6 in 
Vista" howtos on the Internet as well as a number of misguided
people recommending to disable it.

However, the one good thing is that in Vista, as well as in
Windows 7, Microsoft has made IPv6 a requirement for Windows Meeting
Space, as well as Windows Peer to Peer Networking Platform.  This is
a step in the right direction towards getting people to switch
over.

For these reasons, getting the edge fully IPv6 compliant is going to
take solid Windows 7 penetration which won't happen until 2011 at
the least.


Ted