Hello Rob, thanks for your e-mail! Please find my replies inline.
On 28 Jul 2021, at 11:33, Rob PE1CHL via 44Net
<44net(a)mailman.ampr.org> wrote:
On 7/28/21 12:31 AM, Antonios Chariton (daknob) via 44Net wrote:
Fellow radio amateurs, I am writing to you on
behalf of the ARDC TAC, which I represent.
Those of you that were on our Community Call last Saturday may remember that I promised
you we would share our first proposal with the community. A few days after that, I am
happy to send that to you for your review, feedback, comments, questions, and
information!
You can find our 5-page PDF here:
Well, I see a couple of issues with that, which I will detail here:
Even with higher-speed radios, amateur radio-based networks typically will have
limited
bandwidth and higher latency than the public Internet. For instance, HAMNET is an
active
network deployed as a mix of a static, dynamic, and mesh routed network. There can be
many
hops over radio nodes to get to the destination. Each node will add latency and most
links will
have less than 100Mb/s of bandwidth. Considering these conditions, these networks
should not
be used for high bandwidth applications such as consumer video streaming (eg. Netflix,
Hulu).
I think this is an issue that has to be managed locally, not by ARDC. Each local region
has to
decide how much bandwidth they make available in links, and what usage they determine to
be
appropriate. This has to be communicated to the local users. An amateur viewing Netflix
via
AMPRnet is likely not something we would want to see, but I don't think that making
the AMPRnet
an island is the way to achieve that.
You are correct that this is something that should be managed locally and not by ARDC. It
was brought up as an example and the TAC has no intention of setting a policy at this
level. This can be seen from our resolution that does not mention anything related. I
apologize for the confusion.
Routing to amateur radio-based networks needs to be
easy to implement and can be supported by low-cost routers
Users and operators of amateur radio-based networks typically will have routers that
are
connected to both the radio-based network and the Internet and will want to pass
traffic to the
respective networks. This creates a challenge for these users: how to tell the routers
which
network connection should be used, based on the packet destination. If dynamic
routing
protocols such as BGP were used, this would lock out almost all operators as
consumer-grade
routers do not support these protocols. Dynamic routing protocols also require much
more
memory and CPU power than these routers support.
Well, I think that is an issue that we were going to handle with the new backbone
network!
Simple end-users would connect to a PoP and send all the AMPRnet traffic they cannot
route locally
(i.e. to their LAN or to a radio link) towards the PoP and have all the dynamic routing
done in
the new backbone network. That includes routing towards other regions on AMPRnet, and
optionally also routing towards internet. The backbone network would be
internet-connected
in several places and find the optimal place to route that traffic towards internet,
there is no
need for the end-user to bother with that. They only need to make sure they route only
their
AMPRnet traffic there, and not all their normal household internet traffic. But that is
something
that a consumer-grade router cannot do anyway, so they would often need a more suitable
router "behind" the ISP provided router when they would want to route AMPRnet
traffic to internet.
Those users that see AMPRnet as an island can just route 44.0.0.0/9 and 44.128.0.0/10 to
the PoP
and the rest to their ISP, and the typical ISP router can do that. When you really want
to do all
AMPRnet routing on an ISP router, you only need to make sure that your PoP can handle a
VPN
type that the typical ISP router can setup.
Also I do not understand why this is now brought up as an issue, at the moment where we
have
many different types of routers available in the 50euro/$60 price range that can do the
policy
routing and BGP routing. I would not want to advocate making network policy decisions
based
on "it should be usable with static routing" at THIS moment. Maybe 20 years
ago, but not now.
We need to look at the current and future situation and it looks bright w.r.t. dynamic
routing.
We currently have a large number of users that have reached out that have an ISP-provided
router that they do not want to change, regardless of cost, as it’s simple and it just
works. These users have a radio on their roof that connects to an Intranet like HAMNET.
They want a single prefix they can point as a static route to this radio, and the
remaining of it to be caught by the default route they have to their commercial ISP. The
equipment their ISPs typically give them does not support any kind of VPN or tunnel.
Moreover, if they install e.g. a Raspberry Pi to do the VPN, then they still have the
issue of “what prefix do I need for the static route”. The 44 address space that is
"Direct BGP" made the conscious decision of being available on the Internet,
which means that these users should reach them via their commercial ISPs. Some other
operators want to only be reached over the Intranet. This proposal gives a way for all
these users to finally connect to the Intranet that they want to join without forcing them
to massively complicate their network with more equipment, VPNs, dynamic routing
protocols, etc.
We therefore believe that it massively decreases the barrier to enter in both cost,
knowledge, complexity, and time. For the current TAC, lowering the barrier to entry is
something we need to address now, and not in 20 years. We do not like to exclude people
and tell them to come back in 20 years because they’re not as knowledgeable or willing to
spend the time, money, and effort right now. Instead, we try to serve them today, and help
them join the 44 Net.
The TAC analyzed the current allocations within
44.0/9 (USA) and 44.128/10 (RoW). Based on
the existing use-cases within both ranges and the necessary effort to renumber, the
TAC
proposes to use 44.128/10 as the prefix for the radio-based network.
Please show us that analysis, because I arrive at a different conclusion. I regularly
download
the advertised ranges from
http://thyme.rand.apnic.net/current/data-add-ARIN and save
the
AMPRnet portion of that, and at the moment it appears that 182 subnets are advertised
from
44.0.0.0/9 and 152 subnets are advertised from 44.128.0.0/10. Considering that the
first
network is twice the size, the number is comparable or is less for the first than the
second.
However, when looking at the number of advertised IP addresses rather than the number of
networks, the outcome is: 190720 addresses are directly advertised from 44.0.0.0/9 and
587008 addresses are advertised from 44.128.0.0/10, that is three times as much.
So, there is much more effort to renumber when moving those advertised networks in
44.128.0.0/10
towards 44.0.0.0/9 than when doing it the other way around.
It is true that there are more advertised IPv4 addresses in 44.128/10 on the Internet.
However, according to the study that we did, and is linked in the original post, there is
a *massive* discrepancy between advertised space on BGP and actual usage. We see large
parts being advertised (entire /16’s) yet only 2-3 users with < 20 hosts within them.
Although we looked at the allocation plan, we identified that it’s more helpful to either
look at it in more depth (e.g. end-user allocations) or also rely on the ICMP checks we
did, where possible.
These checks showed us that about 75% of the IPv4 addresses that respond to ping are
available on a large Intranet and the vast majority of that is in 44.128/10. That said, if
we did not use 44.128/10 for the Intranet part, we would cause around 75% of the global
users (in terms of IPv4 addresses responding to ping) to renumber.
To sum up, I agree that 44.128/10 has more advertisements on the Internet, but in our
detailed look, the vast majority of hosts within it are Intranet-exclusive. Therefore, we
decided that this should be the Intranet prefix.
Furthermore, the use of the 44.190.0.0/15 network is
currently specificially for networks that
are part of AMPRnet but are advertised as internet services. This was done as an earlier
method
to solve the "problem" that this proposal appears to try to solve, and when now
doing it in the
way of this new proposal that would all have to be reversed again.
This proposal aims to make it as simple as possible for the users to join the network.
This can be achieved with a single static route in their ISP-provided equipment, for
44.128/10. The users don’t want to have to create multiple entries (some equipment limits
to 8 static routes) or have to update them regularly every time we allocate something, and
until everyone changes their static route to have network splits or reachability problems.
If this passes, they can add a single static route (to a radio or a VPN server) and never
have to worry about changes again.
Therefore we believe that this is a holistically better option than what 44.190/15 was.
Finally, I am not in favor of any changes that are
geared towards making the AMPRnet space
a "private, isolated" space. We have the 44.x allocations because they can be
routed to internet,
when people do not want that they can use a 10.x allocation and be perfectly isolated
from the
internet. There is no need to have publicly routable space for that (and then declare it
to be
not routed by policy).
Using any RFC1918 prefix for the Intranet has been considered by the TAC. Unfortunately,
it comes with a set of problems that don’t occur in the currently proposed scheme: because
everyone can use these networks, by definition, there *will* be collisions, and there will
be no central coordination and guarantee of non-overlapping spaces. Nobody can prevent two
“Intranets” from using the same address space, a lot of users already have a 10/8 network
on their homes, that will most likely have overlaps with the Intranet, etc. By using
44.128/10 that is centrally managed and coordinated by ARDC, we guarantee that each user
can enjoy a globally unique allocation that is guaranteed to not overlap with existing
network infrastructure or with other networks they want to interconnect with.
The TAC sees no problem in using “publicly routable space” for the Intranet as it’s the
only type of space that can currently guarantee this property of no overlaps and global
uniqueness. Furthermore, ARDC has currently a very very large number of IPv4 addresses,
and depending on how you count, the usage by end-users is less than 1-5%. I think that
even if you count by /16’s on the Internet, the usage is still less than 5%, despite them
being empty.
As you mention, a lot of TAC members did not like the idea as well as it was a waste, but
since we could not find a better solution, we arrived at a tradeoff: we could either
support many more users and make it much easier for someone to join at the cost of IPv4
space, or we could make it harder to join, require special equipment, etc. but conserve
IPv4 space. This was an easy decision: we want a lot of users, we want to make it easy for
them, we want them to enjoy being part of the network, and we have way more than enough
IPv4 space that we need. And this is why we decided that this approach is better.
With the numbers provided above, we can give every current user an allocation in 44.0/10
and one in 44.128/10 and we would still have less than 10% usage of the overall address
space, which means that we could still support over 8 times as many people and their
networks. And this is a “worst case”: we take the highest usage percentage and we assume
that all users want IPs in both spaces. With conversations we had with some users, we
believe this is far from the truth. We have at least 2x/15 in Germany that only want to be
on the Intranet part.
The proposal seems to consider two disjoint use cases,
one is "the radio network" and the other
is "the internet connected network". That is also reflected in the 44.190
allocation, and likely
is behind many of the current /24 allocations where the user simply wants to play with
servers
on internet, hopefully for a use case related to amateur radio, and is not making a radio
network
at all.
However, what we have here in the Netherlands is a combination of the two. It is a
radio
network, with additional internet tunnels to join areas without radio links between them
(at least
as an interim measure), but also it is internet routed and announced as a /16. We do
not
want to become an intranet!
I think that use case has to remain, and has to be detailed more in the proposal.
We want to be able to accommodate for every use case with our proposal, so we don’t want
to leave any user or community behind. Perhaps it’s better to think of the two spaces as
“Internet” and “Intranet”. The first use allows communication of hosts to and from the
entire Internet, 0/0. The second use case is a private network that only ham radio
operators can access. This can be connected with radio links, satellites, VPNs over the
Internet, etc. We do not want to force you to use a specific technology. We just want to
make sure that we provide a network for both use cases.
Your users in The Netherlands may want to remain connected to HAMNET in Germany and the
rest of Europe. They may also want to have Internet access with publicly routable
addresses. Both these use cases are fine and perfectly acceptable.
If you determine that you need to be connected to both networks, and connection to the
Intranet is not simply enough, you can request a matching allocation (I think /17?) in
44.0/10, and then set up a “netmap”. This is an iptables target (also available in
RouterOS) that replaces the first bits of an IPv4 address. With this, you can leave all
the IPs intact so you can communicate with Germany, and on the single point that you
connect to the Internet you can advertise the new prefix, and perform “netmap” of the
entire old /17 to the entire new /17. So the Internet will see you as 44.0/10 and the
“radio network” / Intranet will see you as 44.128/10.
This is something that is supported by design and we hope it addresses your use case and
provides you with the necessary flexibility to join the network without compromises.
Thanks,
Antonis