[NNagain] are you Bill Woodcock?

[Dick Roy]

Bill,

Thabnks for the very cogent explanation of the realities of IXPs.  Awesome!
If you do end up having that virtual chat, I'd love to be a "fly on the
wall" for that! :):):)

Cheers,

RR

-----Original Message-----
From: Nnagain [mailto:nnagain-bounces at lists.bufferbloat.net] On Behalf Of
Bill Woodcock via Nnagain
Sent: Thursday, January 18, 2024 2:39 PM
To: thejoff at gmail.com; Network Neutrality is back! Let´s make the technical
aspects heard this time!
Cc: Bill Woodcock
Subject: Re: [NNagain] are you Bill Woodcock?

> On Jan 18, 2024, at 22:51, le berger des photons via Nnagain
<nnagain at lists.bufferbloat.net> wrote:
> First I've ever seen the term IXP.  It seems interesting.  Can you point
me to some documentation at a level which only requires the ability to read
in english?  Lots of what I've seen here has initials for things which I
haven't even been able to decode.
> I've been connecting 200 families in a 25 km radius to internet via 8
fiber optic connections for the last 20 years.
> I've been thinking of inviting others to participate,  help them get
going.
> Thinking how it might be useful to provide each client two accesses.  one
to the global internet,  one to a local network which isn't being watched by
big brother.
> Does any of this warrant my looking further into IXP technology?

Hi, Jay.

I’m afraid I’m really bad at getting all this stuff written down, though I
know it would be useful.  I am planning to write a doctoral thesis on
exactly this topic (the societal and economic impact of Internet exchange
points) for Universite Paris 8 next year, but that will need to be a bit
more academic than practical, to satisfy, you know, academia.

So, really basically, it sounds like you’re already building an internet
exchange.  Internet exchanges are where Internet bandwidth comes from.
Internet service providers bring Internet bandwidth from IXPs to the places
where people want to use it: their homes, their offices, their phones.
Internet bandwidth is free _at_ the exchange, but transport costs money.
Speed times distance equals cost.  So the cost of Internet bandwidth is
proportional to the speed and the distance from IXPs.  Plus a profit margin
for the Internet service provider.

So, if one Internet user wants to talk to another Internet user, generally
they hand off their packet to an Internet service provider, who takes it to
an exchange, and hands it off to another Internet service provider, who
delivers it to the second user.  When the second user wants to reply, the
process is reversed, but the two Internet service providers may choose a
different exchange for the hand-off: since each is economically incentivized
to carry the traffic the shortest possible distance (to minimize cost, speed
x distance = cost), the first ISP will always choose the IXP that’s nearest
the first user, for the hand-off, leaving the second ISP a longer distance
to carry the packet.  Then, when their situations are reversed, the second
ISP will choose the IXP nearest the second user, leaving the first ISP to
carry the packet a longer distance.

This only works (and achieves “fairness”) if there’s an IXP near each of the
two users (or they’re both close to the same IXP).  If there’s no IXP close
to the second user, they wind up paying for long-haul transport in both
directions, and their share of the costs are higher than the first user’s.
So, ISPs (and users) are economically incentivized to build small IXPs all
over the place.

IXPs are only maximally effective if they really are free.  If they start
running up costs, which have to be defrayed, and placing the burden of those
costs on the ISPs which try to use them, then they increase the _average per
bit delivery cost_ or APBDC of the bandwidth, making it less affordable, and
causing ISPs to seek more affordable bandwidth elsewhere.  So an effective
IXP is a cheap IXP.  “Gold plating” IXPs kills them.  Fancy is bad, simple
is good.

In the 1990s, there were a diversity of kinds of IXPs
  we were all trying
different experiments, and nobody had settled on a single best way of doing
it yet.  Then, gradually, it all narrowed down to a single most-efficient
model, and all IXPs were an Ethernet switch in a closet, surrounded by BGP
routers which had one port facing the switch, and one or a small number of
ports facing their ISP’s network.  But in the last ten years or so, things
have started to become a little more diverse again, so what you’re doing
would probably be recognized as a form of IXP by many people.

Economic compartmentalization is really important in IXPs.  Some people call
this “neutrality,” but that’s a difficult word to define, because it means
different things to different people.  What’s important is that the IXP is a
shared, communal, enterprise, and the group of parties who are collaborating
to make it go are often business competitors, which means that they need a
very simple system that doesn’t require that they trust each other very
much.  So, if it handles money at all, it’s very hard to get over that trust
threshold. If it has complicated rules or governance, it has difficulty
getting over that trust threshold.  Simple is good.

So, if you’re thinking of making things more complicated (commercial access
(“transit”) to the global Internet), that’s fine, and may be exactly the
right thing to do, under the circumstances
 but you should be very careful
to compartmentalize that, and its finances and risks, into a separate entity
than the fiber, or the “exchange” or whatever else you’re doing.  Otherwise
people will worry that you’re going to use fees from one thing to subsidize
another, which will compete with their interests.  That’s not hypothetical,
that’s actually one of the most common ways IXPs fail: they lose their
neutrality, and lose the trust of their participants, who form a competing
exchange nearby, and all move over to it.

All of this is generally easier to explain in a dialog.  I’m in Paris, so
happy to chat with you on the phone, if your spoken English is better than
my spoken French.

                                -Bill

_______________________________________________
Nnagain mailing list
Nnagain at lists.bufferbloat.net
https://lists.bufferbloat.net/listinfo/nnagain