More Sidgemore on per-bit pricing

Mailman List Mirror (Read Only)
1

Sidgemore talks again:

http://www.news.com/News/Item/0,4,29533,00.html?st.ne.fd.gif.a

I'm sure this subject has been discussed at length here, but I still fail
to see any benefit for per-piece pricing other than to the seller's
revenue stream. I think it's always interesting that discussions of
per-piece pricing always fail to mention the fundamental issue: provide a
disincentive to the consumer to use the service.

There is already per-piece pricing: 1.544Mbps of traffic costs $500 to
$2000+/mo, depending upon how much the provider is counting on you to not
use the service. However, in this case the burden is placed on the ISP to
balance closing the sale with making money off of how much the customer
actually uses the service.

The thing I most distrust about people who talk about per-piece pricing is
that in spite of the fact that it should end up costing lower-use
customers less, for some reason it always ends up costing everyone more.
Look at UUNet's pricing now (for 256K ASU, or something like that), versus
the flat-rate price available from equivalent NSP's. Is that where things
are going?

Pete.

2

One possible positive effect (for the consumer) of "per-bit" pricing
is the opportunity to buy larger pipes but only pay for what you use.

Right now flat-rate pricing mostly assumes you're going to, within
some statistical model, actually use the bandwidth you get, or
certainly that someone buying a DS3 is going to use a lot more
bandwidth, on average, than someone with a DS1.

So if you usually only use, say, 25-75% of a DS1 then you'd be
well-advised to only buy a DS1, even if 10% of the time you really
have a good use for more.

For example, imagine a company which wants to back up all the remote
servers in their branch offices once per week, but would like to get
it done as quickly as possible so it completes in the wee hours when
the databases are quiescent. The rest of the time a DS1 might suffice
for the usual email etc, but for those few hours, for the backups
which run on Sunday night, a full 45mb/s would be useful.

Under the flat-rate model that company would have to either buy a DS3
full-time, for around $50K/mo, or live with the DS1 for around $2K/mo
(or possibly find some equivalent flat-rate option in between such as
a 10mb, but that begs the point, they still can't pay for a T1 when
they need a T1 and a T3 when they need a T3.)

It all depends on the actual pricing models which might arise, which
indeed is an arguable point, but without stating that assumption one
can't argue the more general issue.

For example, if I were to offer you either:

  a) DS1 $2K/mo
  b) DS3 $48K/mo
  c) per-bit DS3 $2K/mo/DS1-equivalent pro-rata per minute

you'd be foolish, unless you have no control over your bandwidth usage
(a whole other issue) not to take (c) if you occasionally need bursts
of high bandwidth and were seriously considering going to the DS3; the
most it can cost would be $48K/mo if you used that bandwidth 7x24
100%, but you almost certainly won't, so it can only cost you less.

Now, it's admittedly not likely that someone wouldn't charge some sort
of premium for the pro-rata bandwidth, perhaps it would actually be
$2500/DS1-equivalent, but unless you really need the DS3 100% of the
time (in which case you should buy the dedicated DS3) it's still
likely to cost you a lot less.

At a 50% utilization that $2500/mo pro-rata still only costs you
around $30K/mo which is significantly less than the $48K/mo for
committing to the whole pipe.

As we move to, e.g., gigabit pipes I suspect this sort of pricing
model will become more and more popular, or else not too many gigabit
pipes will get sold.

Buying 100% of a gigabit at the DS1 equivalent of, say, $2K/mo would
come to about $2 million/gb/month.

Customers certainly exist for $2M/mo connections, but not many of
them.

And the incremental cost of installing a pipe capable of gigabit
bandwidth on demand is relatively very small (mostly just a piece of
fiber and the boxes on each end), likely to be worthwhile even if the
customer just uses $10K or more per month of that potential $2M/mo
(pro-rata), or something like that much, much smaller than the $2M/mo.

We're a case in point of that, we have a good amount of fiber to our
offices here, NYNEX was happy to put lots of it in rather than ever
trench the street again (they had the backhoes out for us on this
urban street twice.)

However, one remaining problem is the local loop, the RBOCs, who do
like to price any service as if you are going to use 100% of it.
Thus, to even buy a per-bit DS3 from an ISP you would still have to
pay my RBOC around $5K-$10K/mo just for the local loop under the
current model, which if nothing else raises the bar for such services.

I think this is something Sidgemore is accounting for in his comments,
however.

3

The other possible effect is that you buy one of these, and then someone
launches a DOS attack at you and you get the bill for it.

The economic impact of this should not be underestimated. Per-bitrate
pricing is a problem as long as the receiver pays for the receipt of
transmissions they may not have solicited.

4

Barry Shein wrote:

One possible positive effect (for the consumer) of "per-bit" pricing
is the opportunity to buy larger pipes but only pay for what you use.

Right now flat-rate pricing mostly assumes you're going to, within
some statistical model, actually use the bandwidth you get, or
certainly that someone buying a DS3 is going to use a lot more
bandwidth, on average, than someone with a DS1.
[Rest of post deleted for brevity]

I wonder if the northeast is more expensive than elsewhere, but from my
recent shopping for T1's for myself and my clients, I find the cost of
the service over a T1 isn't the budget buster. For one location, all T1
circuits (before buying IP service, just the telco charge) is $613 a
month. To another location, the circuits are $900 to $1500 a month.
Adding $500 to $1000 on top of that for full-rate service, vs. adding
$200-$500 on top of that for "burstable" service just doesn't generate
much excitement.

Until the base telco circuit prices are lowered dramatically, the
pricing of packet service over them, while not "noise," is certainly
less interesting.

Now, if the whole circuit, T1 and IP packet service, were all priced on
the basis of traffic, that'd be interesting. An underutilized T1 would
incur some small base charge, plus traffic/usage increments beyond that.
That'd be quite attractive, though I doubt the phone companies would
think so.

Dan

5

Well, a paraphrase of the above is: We must engineer the net to keep
the cost of criminal activities to a minimum so we can continue to
avoid solving the underlying issue.

That's not ridiculous, it may even be an unavoidable factor, but it's
still somewhat sad.

6

Barry Shein wrote:

> >
> > One possible positive effect (for the consumer) of "per-bit" pricing
> > is the opportunity to buy larger pipes but only pay for what you use.
> >
> > Right now flat-rate pricing mostly assumes you're going to, within
> > some statistical model, actually use the bandwidth you get, or
> > certainly that someone buying a DS3 is going to use a lot more
> > bandwidth, on average, than someone with a DS1.
> > [Rest of post deleted for brevity]
>
> I wonder if the northeast is more expensive than elsewhere, but from my
> recent shopping for T1's for myself and my clients, I find the cost of
> the service over a T1 isn't the budget buster. For one location, all T1
> circuits (before buying IP service, just the telco charge) is $613 a
> month. To another location, the circuits are $900 to $1500 a month.
> Adding $500 to $1000 on top of that for full-rate service, vs. adding
> $200-$500 on top of that for "burstable" service just doesn't generate
> much excitement.

I don't think the model becomes very exciting until you model it for
DS3 and beyond, as you say below that there's not enough money
involved to get excited about.

But even at around $1K/mo for a T1 flat-rate, simple extrapolation
puts a 1gb line at $1 million per mo, even if there's a 50% discount
for that, $500K/mo, the customer base would be somewhat limited. Even
a 155Mb/s ATM calculates to around $100K/mo using straight
extrapolation. Being able to get an atm line in for, say, $10K/mo and
then paying the burstable rate might seem very attractive to some.

> Until the base telco circuit prices are lowered dramatically, the
> pricing of packet service over them, while not "noise," is certainly
> less interesting.

This is certainly true, and no doubt something someone in Sidgemore's
position is considering since Worldcom does local loop or is certainly
capable of affecting local-loop pricing.

> Now, if the whole circuit, T1 and IP packet service, were all priced on
> the basis of traffic, that'd be interesting. An underutilized T1 would
> incur some small base charge, plus traffic/usage increments beyond that.
> That'd be quite attractive, though I doubt the phone companies would
> think so.

Exactly my point.

7

Yep.

It is sad.

However, as long as we permit people to source traffic without cost and do
so through proxies, this problem will exist.

This is the primary argument AGAINST anonyminity on the Internet. Your
activities, anonymous or not, are not without cost to others. The entire
premise that you have a right to "anonymous speech" is based upon the fact
that you do not directly harm others economically or otherwise be
exercising it.

However, on the Internet, this is simply not true. "Recipient pays"
is a part of ALL Internet service, and always has been in one fashion
or another - even when the majority of traffic was moved via modems
in the 1980s and early 90s.

Note that this is VERY different from the phone or postal service
networks, both of which are nearly 100% SENDER pays. The exception is
cellular service, and there it is a CRIMINAL ACT to call a cellular
phone on an "unsolicited" basis - that is, to cost-shift where there
is a reasonable probability that the cost is unwanted. Further all
phone traffic is authenticated and can be traced to the source;
"spoofed traffic" (beyond activity which is per-se criminal such as
cloned cellular phones) doesn't exist.

If all transmissions had to be identifyable as to their source, and
chargeback capability was included (ie: if you spam me, I can charge
the transmission back to you - likewise if you ping-flood me) then
the problem would go away. But doing this requires strong authentication
and non-denyability of the transmission itself, which flies in the face
of those who scream for the ability to source anonymous traffic of one
form or another.

That engineering standards have not already stabilized to prohibit
sourcing of traffic with spoofed source addresses, enforced by the
providers themselves, is very much a telling factor here.

There wouldn't BE a DOS problem on the Internet via-a-vis ping floods,
SYN floods, etc. if the provider community refused to permit a connection
to be made without airtight packet source filters which prohibited the
transmission of data with unauthorized source addresses.

Add to that a "chargeback" mechanism (that is, refutation of authorization
for the transmission) and per-bit pricing can work.

Absent BOTH of those on a worldwide basis and I could never justify
recommending to anyone that they accept such a pricing system.

8

Of course you could, if the per-unit cost were the same, pro-rata, as
paying for the whole thing. So if the choice was between paying
$48K/mo for a DS3 vs $2K/mo for each DS1-equivalent the worst case is
$48K/mo anyhow so may as well take your chances with crooks.

I'd argue that even at some premium, unless you're normally using
almost all of it, you'd still probably be better off on average even
if a crook gives you a bad week once in a while. $24K + $6K for a
crook is still a lot less than $48K guaranteed, crooks or not. Etc.

And there's probably going to have to be some accomodation (e.g.,
credits, refunds, more vigorous enforcement and detection of abuse by
sellers) with any such pay-per-bit scheme. Particularly if, as I
predict, it becomes a major way to sell a lot of very high bandwidth
lines (155mb+) to customers who otherwise wouldn't consider so much
bandwidth if they had to pay for all of it all the time.

You're right that something has to be done, but I don't particularly
accept that the situation is so untenable. On a service like this a
credit for a bad week with a crook doesn't really drive the provider
under either, particularly if they make some effort to prevent it
(e.g. prosecuting abusers, detecting and blocking abuse quickly, etc.)

I'd guess that one model which might work well is whitelisting: I want
on-demand bandwidth up to, say, 155Mb/s to this short list of sites
(VPN-ish), but only T1 to everyone else to prevent abuse.

9

One notable thing about per-bit pricing, as well, is that once one
provider really rolls it out and pushes it at a low cost, it's bound
to become a surety. Suddenly the provider with per-bit billing will
be able to steal all of the low-usage customers, while leaving the
expensive near-saturation customers at other providers... effectively
raising their cost per DS1...

As to the merits of per-bit pricing.. I could certainly see myself
purchasing a lot more bandwidth if I could use it on demand, even
with the possibility of someone deciding to smurf me. And think of
other positive effects-- there will be real economic forces urging
customers of providers not to be smurf relays, etc.

Mike

10

However, as long as we permit people to source traffic without cost and do
so through proxies, this problem will exist.

This is the primary argument AGAINST anonyminity on the Internet. Your
activities, anonymous or not, are not without cost to others. The entire
premise that you have a right to "anonymous speech" is based upon the fact
that you do not directly harm others economically or otherwise be
exercising it.

However, on the Internet, this is simply not true. "Recipient pays"
is a part of ALL Internet service, and always has been in one fashion
or another - even when the majority of traffic was moved via modems
in the 1980s and early 90s.

Note that this is VERY different from the phone or postal service
networks, both of which are nearly 100% SENDER pays. The exception is
cellular service, and there it is a CRIMINAL ACT to call a cellular
phone on an "unsolicited" basis - that is, to cost-shift where there
is a reasonable probability that the cost is unwanted. Further all
phone traffic is authenticated and can be traced to the source;
"spoofed traffic" (beyond activity which is per-se criminal such as
cloned cellular phones) doesn't exist.

If all transmissions had to be identifyable as to their source, and
chargeback capability was included (ie: if you spam me, I can charge
the transmission back to you - likewise if you ping-flood me) then
the problem would go away. But doing this requires strong authentication
and non-denyability of the transmission itself, which flies in the face
of those who scream for the ability to source anonymous traffic of one
form or another.

That engineering standards have not already stabilized to prohibit
sourcing of traffic with spoofed source addresses, enforced by the
providers themselves, is very much a telling factor here.

There wouldn't BE a DOS problem on the Internet via-a-vis ping floods,
SYN floods, etc. if the provider community refused to permit a connection
to be made without airtight packet source filters which prohibited the
transmission of data with unauthorized source addresses.

Add to that a "chargeback" mechanism (that is, refutation of authorization
for the transmission) and per-bit pricing can work.

Absent BOTH of those on a worldwide basis and I could never justify
recommending to anyone that they accept such a pricing system.

Those price mechanisms are possible on connection-oriented networks, such as
X.25 and ATM networks.
On connection-less networks such as IP networks, the source will always have
the right to send traffic; packet filtering and traffic shaping can cut some
of the possibly unwanted traffic, but not all of them.

RUbens Kuhl Jr.

11

One notable thing about per-bit pricing, as well, is that once one
provider really rolls it out and pushes it at a low cost, it's bound
to become a surety. Suddenly the provider with per-bit billing will
be able to steal all of the low-usage customers, while leaving the
expensive near-saturation customers at other providers... effectively
raising their cost per DS1...

I have seen it argued several times that if the price is "low enough",
customers prefer a fixed price, even if that price is slightly higher
than they could get by a variable-rate pricing scheme.

Why do you think per-bit pricing would be significantly different here?

Steinar Haug, Nethelp consulting, sthaug@nethelp.no

12

Well, if you ever want to download DVD's or (HD DVD's in the future)
in anything approaching realtime or less then we need some way to
make 10-100Mbit pipes into homes economical.

Current high bandwidth pricing ($100K/month or so for OC3 Internet
connectivity) is just ridicuolous and in no way related to the
value that is created for the customer (you can argue that point
since some people are paying...)

Computers have gotten faster by a factor of x over the last n years
yet bandwidth costs have hardly changed in the last 10 years. This is
despite the fact that the capacity of fiber already in the ground
keeps increasing by leaps and bounds. Somehow this problems needs
to be solved.

If some sort of usage based pricing doesn't do it then we'll have to
wait for super high speed, auto configuring wirelss networking in
each PC. Everybody has a 100Mbit connection to all PC's within
"earshot". The higher the local computer density, the higher the
available bandwidth (with some upper limit of course based on how
much frequency space gets allocated to this). Problem solved,
phone companies roll over and dies.

Dirk

13

> Absent BOTH of those on a worldwide basis and I could never justify
> recommending to anyone that they accept such a pricing system.

Of course you could, if the per-unit cost were the same, pro-rata, as
paying for the whole thing. So if the choice was between paying
$48K/mo for a DS3 vs $2K/mo for each DS1-equivalent the worst case is
$48K/mo anyhow so may as well take your chances with crooks.

Except that if I don't need a DS-3 often then the possibility of being
billed for it when I wasn't the requestor is a hell of a liability.

Particularly if, as I
predict, it becomes a major way to sell a lot of very high bandwidth
lines (155mb+) to customers who otherwise wouldn't consider so much
bandwidth if they had to pay for all of it all the time.

Again, it depends on the risk factors.

You're right that something has to be done, but I don't particularly
accept that the situation is so untenable. On a service like this a
credit for a bad week with a crook doesn't really drive the provider
under either, particularly if they make some effort to prevent it
(e.g. prosecuting abusers, detecting and blocking abuse quickly, etc.)

I'd guess that one model which might work well is whitelisting: I want
on-demand bandwidth up to, say, 155Mb/s to this short list of sites
(VPN-ish), but only T1 to everyone else to prevent abuse.

Possibly, yes.

14

I know this is a bit late sorry...

Well what if I had a xDSL local loop from someone like Northpoint for
something like $150 and then did the per bit model. A T1 (or even slightly
less) for $350 starts to look rather attractive to a lot of people.

Barry Shein wrote:

One possible positive effect (for the consumer) of "per-bit" pricing
is the opportunity to buy larger pipes but only pay for what you use.

Right now flat-rate pricing mostly assumes you're going to, within
some statistical model, actually use the bandwidth you get, or
certainly that someone buying a DS3 is going to use a lot more
bandwidth, on average, than someone with a DS1.
[Rest of post deleted for brevity]

I wonder if the northeast is more expensive than elsewhere, but from my
recent shopping for T1's for myself and my clients, I find the cost of
the service over a T1 isn't the budget buster. For one location, all T1
circuits (before buying IP service, just the telco charge) is $613 a
month. To another location, the circuits are $900 to $1500 a month.
Adding $500 to $1000 on top of that for full-rate service, vs. adding
$200-$500 on top of that for "burstable" service just doesn't generate
much excitement.

Until the base telco circuit prices are lowered dramatically, the
pricing of packet service over them, while not "noise," is certainly
less interesting.

Now, if the whole circuit, T1 and IP packet service, were all priced on
the basis of traffic, that'd be interesting. An underutilized T1 would
incur some small base charge, plus traffic/usage increments beyond that.
That'd be quite attractive, though I doubt the phone companies would
think so.

Dan

--
-----------------------------------------------------------------
Daniel Senie dts@senie.com
Amaranth Networks Inc. http://www.amaranthnetworks.com

Thank you,
David Diaz
Chief Technical Officer
Netrail, Inc

email: davediaz@netrail.net
pager: 888-576-1018
office: 888-NETRAIL
Fax: 404 522-2191

Colo facilities: Atlanta-NAP, Miami, Arlington, Chicago, San Francisco

15

True, but if you become a CLEC and use something like access method 5 you
can get that loop cost to around $30 and also offer voice services using
the first 4 kHz of the loop.

16

Most small ISPs cant afford to do that. Becoming a CLEC is not a $100
turnkey option. However you can get good pricing from many CLECs and then
take a ATM (or even frame) T1 handoff. Pretty small barrier to entry and
it let's you scale from there without having to have the equipment, lawyers
and patience.

I'm also thinking where this model might work well is in bldings where you
can go in, drop a TNTdsl in the basement. Use existing copper to provide
1.5meg (soon 2.23meg/sec??) handoffs to each suite. Then run a frame
through the local RBOCs cloud to your hub. This would allow you to start
small at several bldings and scale from there.

It would make sense to charge by the bit since that most likely will be
your #1 cost. All the equipment can be leased including CPE side which
would be some cheap PipeDSLs.

I know this is a bit late sorry...

Well what if I had a xDSL local loop from someone like Northpoint for
something like $150 and then did the per bit model. A T1 (or even slightly
less) for $350 starts to look rather attractive to a lot of people.

True, but if you become a CLEC and use something like access method 5 you
can get that loop cost to around $30 and also offer voice services using
the first 4 kHz of the loop.

--
Check out the new CLEC mailing list at http://www.robotics.net/clec

<>

Nathan Stratton Telecom & ISP Consulting
www.robotics.net nathan@robotics.net

Thank you,
David Diaz
Chief Technical Officer
Netrail, Inc

email: davediaz@netrail.net
pager: 888-576-1018
office: 888-NETRAIL
Fax: 404 522-2191

Colo facilities: Atlanta-NAP, Miami, Arlington, Chicago, San Francisco

17

Yeah... but you _STILL_ can't get that large a router port for that
little money, and you never will.

Bandwidth is what costs money, not copper.

Cheers,
-- jr 'and people _STILL_ miss this' a

18

Most small ISPs cant afford to do that. Becoming a CLEC is not a $100
turnkey option. However you can get good pricing from many CLECs and then
take a ATM (or even frame) T1 handoff. Pretty small barrier to entry and
it let's you scale from there without having to have the equipment, lawyers
and patience.

That is just it, that is why I said using access method 5, you dont need
to colo in the CO, most ISPs can afford to become CLECs if they can get
rid of that 100 - 300K expense. It is not a $100 turnkey option, but it is
something you could do for 5K. You say that you can get good pricing from
the CLECs, I just have not found that to be true. Have you looked at Covad
and Northpoints pricing? It is hard to make any money at that when the
ISP down the streets cost is $30.

I'm also thinking where this model might work well is in bldings where you
can go in, drop a TNTdsl in the basement. Use existing copper to provide
1.5meg (soon 2.23meg/sec??) handoffs to each suite. Then run a frame
through the local RBOCs cloud to your hub. This would allow you to start
small at several bldings and scale from there.

If you were in the building you could do VDSL and give everybody 53 meg.

It would make sense to charge by the bit since that most likely will be
your #1 cost. All the equipment can be leased including CPE side which
would be some cheap PipeDSLs.

There also is CPE comming out with voice, video, and ethernet jacks all
under $750.

19

I was just saying how you could lower the cost of the ADSL loop from
around $150 to $30, with both you would need to supply the IP. The idea
would be that you would not just use the port for internet traffice. You
would provide voice services, you would provice VOD (video on demand), and
cable services (using something like I Magic TV).

20

Nathan Stratton wrote:

> I know this is a bit late sorry...
>
> Well what if I had a xDSL local loop from someone like Northpoint for
> something like $150 and then did the per bit model. A T1 (or even slightly
> less) for $350 starts to look rather attractive to a lot of people.

True, but if you become a CLEC and use something like access method 5 you
can get that loop cost to around $30 and also offer voice services using
the first 4 kHz of the loop.

Unfortunately you'd still be at the mercy of the RBOC to actually take
care of your line when it breaks. I'm not even sure being a really big
CLEC would help in my neck of the woods.

What I wouldn't give to be able to buy connectivity from someone other
than Bell Atlantic (he says, after waiting 27 hours for a blown mid-span
repeater to be diagnosed and replaced).

Dan