[nanog] Re: Microsoft offering xDSL access

Mailman List Mirror (Read Only)
1

Now, if you want to *CLAIM* DS1 speeds but actually deliver something that
looks more like an ISDN connection, then its possible.

T1 into a local provider is still T1, though. The main bandwidth consumers
hereabouts are HTTP, NNTP, and DNS. All of these are "cacheable" at the
local provider, and having T1 access to those data repositories is still
quite useful. Not all traffic has to go to the other side of the planet.

                                                        But where I come
from advertising something you can't deliver is commonly known as fraud.

We just call that "marketing" around here. But your point is valid: this
low-cost last-mile service should be described in terms of access to the
ISP's local resources (which will therefore have to grow a lot from now).
Hell, maybe there's something in this transparent caching idea after all?

Nobody in their right mind would sell T1 to every customer and do 1000:1
overcommit of their transit links without some kind of bandwidth shaping.
Since that shaping is part of the engineering plan, the marketroids will
have to find some way to come clean about it in the advertising glossies.

2

Nobody in their right mind would sell T1 to every customer and do 1000:1
overcommit of their transit links without some kind of bandwidth shaping.
Since that shaping is part of the engineering plan, the marketroids will
have to find some way to come clean about it in the advertising glossies.

We've got 40 T1 ADSL customers coming into a single 1.1Mb SMDS connection

Traffic is around %3 to %5 on average with occasional spikes to %50 or
%100 for short periods. Im guessing this behaviour is due to a single PC
at the customers side -- limitations on how much traffic a single user can
generate - how much bandwidth a single stream uses. With a DS3 you could
get perhaps 5,000 to 10,000 "T1" customers is a rough guess. Look at
@Home's model for example.

  Stb

3

> Now, if you want to *CLAIM* DS1 speeds but actually deliver something that
> looks more like an ISDN connection, then its possible.

T1 into a local provider is still T1, though. The main bandwidth consumers
hereabouts are HTTP, NNTP, and DNS. All of these are "cacheable" at the
local provider, and having T1 access to those data repositories is still
quite useful. Not all traffic has to go to the other side of the planet.

Well, yes and no.

NNTP absolutely is.

DNS is low-volume enough that its irrelavent.

HTTP is not nearly as cacheable as you would think, and caching it has some
bad side effects in many cases - which your customers will likely bitch
about.

> But where I come
> from advertising something you can't deliver is commonly known as fraud.

We just call that "marketing" around here. But your point is valid: this
low-cost last-mile service should be described in terms of access to the
ISP's local resources (which will therefore have to grow a lot from now).
Hell, maybe there's something in this transparent caching idea after all?

Let's say that you can cache 50% of the HTTP traffic, which frankly, from
what I've seen is HIGHLY aggressive, but I'll be nice and give you that for
the sake of argument.

Nobody in their right mind would sell T1 to every customer and do 1000:1
overcommit of their transit links without some kind of bandwidth shaping.
Since that shaping is part of the engineering plan, the marketroids will
have to find some way to come clean about it in the advertising glossies.

Ok, so its only 500:1 assuming 50% effectiveness on the HTTP side.

It still won't work.

Now, if you intend to rate-shape (as opposed to tossing packets on the floor
when you get overcommitted) then you ARE committing fraud if you don't tell
the truth about it. And, frankly, the customer really gets hosed with this
kind of model - because you have to be pretty predictive for this to give
you any kind of net gain in effective utilization, which means you apply the
chokes BEFORE the peak levels get hit.

The business models on which this stuff is based are full of assumptions
that anyone who has been in this business for more than a week knows won't
work in the real world.

At least, not in the world that we call the "Internet".

Now if you want to set up a private network which is called and is in fact
something less than that (ie: "Citynet"), you might be able to make it work.

Maybe.

The people running around shouting that "bandwidth will be too cheap to
meter" sound a LOT like the crowd that was doing this in relationship to
Nuclear Power and electricity 20 years ago.

Today, the TRUTH is that Nuclear energy is one of the most expensive ways
to generate electricity, and in fact its SO expensive that places like
Illinois, where ComEd generates a lot of it from that form, don't have
enough on peak demand days and have to pipe it in and/or start turning
people off!

4

And I'll bet they're all at different speeds as well, both TX and RX.
It sounds like you have to play it by ear, that coming up with some
mathematical model is *almost* impossible.

Regards,

5

Funny, last year in Boston I spoke to a rep from Highway1
(Media1's Internet division) wherein he stated that they planned to sell
off 2,000 cable modem links (subscribers) from a single T3 while
promising 10Mbps speeds to the home. I pointed out this was an oversale
ratio of something like 400:1. He countered that I must not understand
the statistics of network traffic. I countered that he must be engaged
in wishful thinking if he expected end users not to push the limits of
technology and expect the moon for only $19.95 per month. I predicted
that people would decide that CUCEEME (sp?) was too cool to live without
and that they would want to listen to their neighbor's CD audio over the
'net. Needless to say, we parted company with neither having been
convinced by the other. So in general the cable companies seem to
believe in high oversale ratios, although not perhaps in the thousands.

          David

6

You can't shape to a model you can't develop, but if you're
already established, with those 40 customers you can develop a
bandwidth-distribution layout and rather simply predict trends..

If you are in a highly volatile market, which xDSL doesn't seem to be
quite yet :), this would be impossible, but adding customers at a
constant rate with no changing technology would imply a constant trend..

Of course, as soon as technology changes that is no longer valid :),
but.. alter the model on a monthly basis? ..

I wonder if RX and TX equalize at some point? If you have enough static
[customers] machines connected to be transmitting data at a steady level?
I'd be curious to see some numbers on that..

-greg

7

Fine, but if I can get a "T-1" for $50 a month, I'm gonna put it in
every apartment building and condo in town and make a lot of money.
Until the "T-1" turns out not to be, at which point I'll be joining the
class action lawsuit.

Cheers,
-- jra

8

The business models on which this stuff is based are full of assumptions
that anyone who has been in this business for more than a week knows won't
work in the real world.

While most business models require a feedback loop that is frequently
missing in Internet businesses (assume traffic characteristics, measure
network, correct business model) all of these kinds of tiered service
depend on classes of customers and their usage patterns. If I have classes
of customer that uses a T1 service differently, it is completely
reasonable to sell it differently. For example, if I have a 10 Mb/s cable
modem, but *on average* use it like a 28.8kb/s modem, then why should I
pay for a full 10 Mb/s of bits that I am not sending/receiving? If I get
the 10 Mb/s access but it is engineered to deliver 56k based on my
customer class useage patterns, is that fraud? I think not. What if I
advertise it as 100 times faster than a standard modem? Still not fraud.
Now, if I as a residential cable modem user set up a pay per view web
site that makes cool site of the world, I am do not fit the class of
customer profile. The key issue in all this is what the provider will do
with outliers, and if the prediction of outliers matches the experience of
outliers.

Another way to look at this is that it is in effect a pay per bit system,
that, combined with customer useage assumptions, gives a a tiered service
pricing offered over one type of access. Most ISPs do not price their T1 service
assuming 100% useage 100% of the time. $35/mo over a T1/ADSL/Cable modem
is just an extention of this practice with different assumptions.

Eric Carroll eric.carroll@acm.org
Tekton Internet Associates