UUNET Press Release on Peering

Tue May 13 02:01:53 UTC 1997

On Mon, 12 May 1997, Vadim Antonov wrote:

> On Mon, 12 May 1997, Christopher Morrell wrote:
> 
> >> Which of UUNET's peers who are able to continue to peer with them, would
> >> you say do not follow the criteria set out in the press release?
> 
> >I know of at least one.  I'm sure there are more.  It seems,
> >that equity in traffic exchange may be an overriding
> >principle, although running many web-farms and small
> >backbone can certainly give you that.
> 
> This is a serious misunderstanding of the economics of Internet.
> 
> The "equity in traffic" is not important at all.
> 
> What is important is equal share in carrying packet flows, which
> can be approximately measured in miles*bps.  *That* costs money.
> 
> Therefore, a small ISP dumping most of its traffic to others at
> one exchange point will cause long-haul backbones to carry its
> traffic most of the distance (no matter which direction the traffic
> goes), thus effectively subsidizing that small ISP.

[Warning, arm-chair telecommunications guy is making assumptions below...]

This is a good point. In the LEC market, it is assumed for the most part
that (1) the costs of building infrastructure are pretty much the same for
the ILEC as they are for the CLEC and (2) whether someone is a customer of
the ILEC or a CLEC doesn't matter, because one of them would have had to
incur the costs required to bring that customer on. Therefore,
interconnect agreements amongst the ILEC and CLECs usually are "free".

However, there is something of a balance. First, the LEC where the call
originates is responsible to provide the facilities to the LEC where the
call terminates. This means that the ILEC will for the most part be
providing a lot of service to the CLECs. Remember, though, that the ILEC
would have had to build up capacity anyways, to terminate the call within
their own network.

In the case of a national backbone, I think there are huge differences. 
The NSP incurs huge costs to get to the exchange point, whereas the ISP
(ie Web farm, local provider, etc) really incurs a very small expense (by
comparison). In the case of a Web farm, it can easily increase it's
in/out-ward bandwidth with a comparatively insignificant cost, compared to
the investment required by the NSP to support the farm's increased
traffic.

There may be some other interesting things to consider, especially in the
case of UUNet. One thing I don't really understand is how these
"free-loading" peers are detrimental. If UUNet customers pay on a tiered
pricing scale, that means that when they go to a UUNet peer, they are
paying for the bandwidth being used. What UUNet is saying is that *two*
people should be paying for that bandwidth. 

And how is it any different if UUNet loses the second person's revenue
because the traffic ended up on a "small" peer or a "large" peer. In any
case, it is revenue that's lost, period. Wouldn't the benefits of being
able to manage your network (because you can tell peers where to peer with
you, and anticipate that they'll be peered there a long time) outweigh to
a large degree the shortcomings of a small peer (who, for example, could
buy service from another NSP, who might regularly move traffic around
enough that a national backbone would be significantly affected)? 

I just don't see that many of the small peers will end up buying UUNet
service (especially at their special "wholesale" prices), but will
ultimately end up buying from another NSP, and UUNet will end up not only
not getting new revenue, but will have compounded problems from the same
traffic they had to support before. 

I would also assume that there are some economies of scale by peering at a
central point, rather than putting high-speed customers on POPs away from
the exchange point. If one of these small peers decided to buy UUNet
service, wouldn't it be significantly more expensive to haul DS-3/OC-3
bandwidth close to the customer, whereas they used to haul it to the
exchange point themselves?

Just my $0.02 (well, an email this long is probably worth $2.02).

Pete Kruckenberg
pete at inquo.net