On 10/5/19 11:53 AM, Dave Taht wrote:
Lee Howard <[email protected]> writes:
On 10/4/19 4:55 PM, Dave Taht wrote:
not being able to get a
static IPv6 address out of comcast, my hurricane tunnel getting blocked
by netflix, the still-huge prefix sub-distribution problem. The idea of
dynamic 2 week prefixes in part of the world prone to earthquakes
doesn't work for me...
I can think of several programmatic ways to deal with that.
...
for real use a static ipv6/48 to distribute is needed. Dynamic ipv6
assignments are fine if you are doing trivial stuff but if ipv6 is ever
to even start to supplant ipv4 it's got to become more static.
Or more resilient to renumbering. I didn't mean to suggest that you as a
user should have to code your home network. I meant that we (maybe IETF)
have not done well at handling the case of "I received a new prefix, and
I am the edge router; I need to tell everything else what changed." Thus
kicking off RAs, DHCPv6 updates, dDNS updates, and a firewall that
understands dynamic prefixing.
I was thinking about some Hackathon projects to add IPv6 capability to
open source projects. Seems to me the hardest part is making sure
there's an adequate test environment.
Hackathons ARE useful tools for getting a short burst of focused work
out of people sharing the same space and time, but too many are thinking
hackathons alone will solve more detailed design, coding and iteration
problems; it's one of those ideas trivializing the costs of "Real
Programming(tm)" that really bugs me nowadays.
I could share here the detailed project management stuff that went into
cerowrt's run (3 years), or the outline of work we did for
make-wifi-fast - which we've now been at for over 5 years now - 3+ to
get fq_codel to work right on wifi, 2 to rework the API to work for more
devices, and a pointer to the latest work which has been going for 3+
months now - and for all that we've only accomplished about 1/10th what
we wanted to do, and only on 4 chipsets (most recently intel's ax200
chips) out of the hundreds.
This might make for an interesting WG discussion, getting actual work
organized and done.
But Mr.Rey's reference about IPv6 deployment rates also makes a good point!
Nobody cares about deployment rates. What good does it do, if people don't use
it ?
This is more realistic : https://www.google.com/intl/en/ipv6/statistics.html
During the week, we are below 25%.
(Replying to an item upthread)
APNIC's statistics show that in almost every network that has IPv6, it
is almost always used.
I pointed to coffee shops as one counter example.
If they don't have IPv6, they're not using it.
To the lack of
DHCPv6-PD on android (and I think, IOS) for tethering as another.
Based on discussions at IETF, I guess Android is expecting to get
multiple IPv6 addresses and reassign as needed. Don't know about IOS.
Another thought I've had:
One of the reasons small ISPs can't deploy IPv6 is that they don't
control the features in the CPE, because they don't buy enough.
I know a couple CPE vendors who would be happy to provide a specific
feature set for a guaranteed purchase of a couple thousand units a
month. This sounds like a good business to me: if a bunch of small
ISPs each contract for a specific number of units, but require
RIPE-554, RFC7084, and RFC8085, we could both get the needed features,
and get a larger volume discount than they get now.
Yes, the smaller ISPs should join together in a buying
club like that. Tried to get that going in NZ once. Failed.
tried harder to make the aftermarket do the right thing - the eeros and
google wifi's of the world are doing ok, the bottom part of the market
just copy/pastes whatever's in openwrt at that moment, slaps a label
on it and ships it. So we focused on making the openwrt base as good
as possible.
Might be another part of "What the WG should do" discussion.
Saving $1 per CPE is better than spending $20 for an IPv4 address for
every new user. Please confirm my math. :)
I always thought that ISPs would invest in their CPE far more than they
have. Free.fr being a shining example! ISPs get paid for modem rentals
and have customer support costs that could be reduced - that should have
been a great ongoing funding source and motivation, alone.
but I know a few vendors, like evenroute, doing bufferbloat AND ipv6
right, that have totally failed to crack ISP market thus far.
and for no reason I can think of, the rental folk don't push out
new hardware OR new software to their users - I think charter made
an effort to get docsis 3.1 stuff out there and retire all the docsis
2.0 gear in place, but not comcast.
Secondly none of those ipv6 standards help when you still really need a
real IPv4 address, so yer still out the $20, IF you can buy the /24s you
need. And there's more ipv6 RFCs left without running, integrated code,
to support them.
In my experience, accountants run the CPE purchasing, and saving 5ยข per
unit is worth more to them than avoiding 10% of them driving $30 support
calls. Same with rentals: why replace when the cost is already
completely written off? There are some good reasons (like technical debt
dragging on forward competitiveness) to replace.
It may be the case that not everyone needs a unique IPv4 address.
Several broadband ISPs have deployed DS-Lite, for example. I don't know
the business model; do they charge for unique IPv4 addresses when
needed? That's sort of back to my other point of aligning expenses and
revenues.
You just mentioned your un-upgradable "OS/2 Warp, MS-DOS, Windows 95,
HPUX, Solaris, Windows 2000," and now you say it's easy to upgrade.
I didn't say it was "easy to upgrade" in the context of this legacy
gear, I said it was easy to "add" 420m addresses. 240/4 is almost fully
enabled in every OS except windows, for example. Fixed the last bug
in it for linux and openwrt last december. Deploying. 0/8 now.
What's the relative level of support for unicast 240/4 and IPv6?
Yea, people keep missing on this point. IPv6 is not globally reachable
either. To try and clarify: A new IOT device trying to backhaul its
data to 240.0.0.1 doesn't need to have a windows OS also trying to get to
that same address. Is that clearer? A new application can try to use
new IPv4 addresses.
Do you mean 240/4 for private connectivity, or 240/4 publicly routed and
it's fine for some applications if it's unreachable? Do all routers etc
support it?
Lee
