7 Feb
2021
7 Feb
'21
12:03 a.m.
On 2/6/21 2:59 PM, Jason McCormick via 44Net wrote:
I don't see how deploying some actual hardware is
unrealistic. A VPS-based solution will not scale well because in the VPS, you're
buying into a system that's oversubscribed by design and you're at the mercy of
the software layer inside of the VPS system the host is using. The other consideration is
that right now, practical use of 44Net address space is limited by all of the
technological hurdles you already summarized so well. If we deploy an easy-to-use system
that support things like a single OpenVPN connection for an endpoint or easier tunneling,
then usage is likely to ramp up quickly. Having real systems designed in an
"enterprise" way actually cuts down on user support problems later on or the
need to redesign-on-the-fly as you grow. No one wants to be trying to rebuild the plane
while it's in the air.
One of our problems is that we have a hard time coming up with a use case and
killer advantage of our own network compared to the normal internet.
One of the few hard advantages we have is the availability of large numbers of
IPv4 addresses for free. That enables us to run a network with clean routing and
subnetting and without NAT. Once we would migrate to IPv6 we would also lose
that advantage.
So, there really is no large influx of new users to be expected. After all, why would
you connect your system to 44Net unless you already are a networking buff that
would like to run services at home and play with networking. Not so many radio
amateurs are. In our country with ~13000 registered amateurs (and pop. 17M)
we now have issues 244 OpenVPN certificates over the 6.5 year period that service
is now up and running, and about 25 of them are usually connected.
We also have about 125 AS numbers for routers around the country, and about
170 subnet routes being distributed via BGP (plus those for the VPNs).
That is a scale where software solutions like Linux VPSes either bare or with a router
oriented OS+Configuration like MikroTik RouterOS still works perfectly fine. We are
not to be compared with an ISP with millions of subscribers.
Actually using VPSes instead of hardware makes it easier to ramp up (and down)
quickly, as those providers normally have very convenient mechanisms to deploy
new instances, which beat truck-hauling hardware around the world every time.
Also when you use VPSes from an external supplier, it is them who are looking after
the hardware, replace the broken fans and disks, and replace the entire machines
when they become obsolete.
I think the only work we would need to do ourselves is to make a network design
and base configuration of the equipment, and likely make some tools to assist
in chores like adding a user account or a new node in the network.
(doing reconfiguration for that)
Some people come up with the "disadvantage" that relying on a VPS provider
would
mean that your network is less reliable because you rely on some unknown external
party. I think it is not so bad because we can always decide to use VPSes from
different providers, and even when you provide your own hardware you will still
rely on others like a datacenter with their power, cooling and networking infrastructure.
Plus you then need to support your hardware or have people do that.
I don't think a small group of radio amateurs should think they can do a better job
than people like AWS, OVH, LINODE, Microsoft Azure, etc which have dedicated
professionals working 24/7 to maintain, monitor and improve their infrastructure.
Rob