Pay-As-You-Use High-Speed Internet?

Mailman List Mirror (Read Only)
1

Hello Fellow NANOG'ers,

I was just thinking about this - tell me if it sounds reasonable? The company that I work for developed a piece of technology which, through rate-limit statements, allow customers to buy/sell bandwidth "on demand". Now, I was thinking: "Why can't we take this technology that we've tested successfully in a colo environment and adapt it a little bit for personal/buisness-class ISP's to allow them to bill for the bandwidth that a customer uses, and only that with the exception of a base monthly fee (to cover the DSL/T1 loop, e-mail services, support, etc.) of a few dollars.

Personally, I would like to see a senario where everyone just pays for what they use - it would be a much better system for allowing people who don't neccessarily need to get on the Internet at high-speed, get on high-speed which will not only increase revenue for the ISP's, but also for the customer who can now use DSL/T1 access in a much more effective way.

Questions? Comments? Suggestions?

-- Jonathan

2

Who pays for a DDoS attack, or getting flooded by bounces from a spammer's
joe-job or A/V companies warning spam when somebody else's box spoofs my
e-mail address?

If they have a website, who pays how much if it's slashdotted? (Serious
question there - I may have budgeted for only several hundred or a thousand
hits a day, and if 200K hits costs too much, I may be in trouble...)

How do you handle disputes? Who has the burden of proof?

Those are all questions I'd be asking as a potential customer..

And the biggie for you is: How do you handle these issues on a low margin? :wink:

3

To answer your question, in our colo evironment, incomming traffic is free and not measured for billing purposes (but I assume this will be different on the ISP platform).

As far as being slashdotted, if it does happen - then your agent from our application will watch - and adhere to - the budget that you had initially set and any "Quick Response" settings that you had set, too.

Disputes, as far as what? The bandwidth that is purchased is all logged into a database for review/auditing. As for the burden of proof, see my previous statement.

-- Jonathan

4

In an application where you pay-as-you-go with hard limits, the site stops responding under the slashdotted activity. The limit protects the ISP and the customer from a dispute, and the customer decides whether to rethink their hard limits or the popularity of their content.

DJ

Jonathan M. Slivko wrote:

5

Woops.... Almost forgot to answer the most important question:

> And the biggie for you is: How do you handle these issues on a low margin? :wink:

Well, to answer that question, it really doesn't take that much work for us, as we would only be licensing our technology to the ISP, we wouldn't be the end-users ISP, only a mechanism for doing this mode of billing/access methodology.

6

Agreed.
-- Jonathan

Deepak Jain wrote:

7

Well - you could, to save costs, put a T3 (or multiple T3's) into a specific area that you want to serve and then distribute it from there via Ethernet. This is what we're currently doing with a residential/commercial building.

-- Jonathan

Daniel Senie wrote:

8

Also, you could also take the approach of wiring a whole building for Internet connectivity through that model, like Intellispace does.

-- Jonathan

Daniel Senie wrote:

9

Okay, so basically, I'm in complete sympathy with you, because I would
_like_ the overhead cost of an unutilized local loop to be zero.
Unfortunately, that's not the case in the not-entirely-ideal world in
which we live.

Also, the idea you're bringing back from the many-times-dead lies in the
shadow of the invisible hand of Enron. Which is unfortunate, since at
heart, it's a good idea.

                                -Bill

10

Bill - I'm not saying dedicate a whole T1 to a single customer, i'm saying share a T1 or T3 among many customers in a small geographic area, but let each customer have fair use of the T1/T3.

BTW, we have been doing this for the last 6 years in a colo environment and more recently a residential/corporate building with about 300 units (50 of which are lit by us) with a single T1.

As far as the local loop cost being zero, I *know* that that is not feasible, but what is feasible is to make a fixed cost aside from the bandwidth of say $30-$50 per customer per month to cover the cost of e-mail service, support, etc.

-- Jonathan

Bill Woodcock wrote:

11

For an idea to catch on, it often helps for there to be a clear benefit to
doing things the new way rather than the old way (or at least, it needs
some good marketing...).

In this case, it's not clear to me where the benefit is. A lot of the
cost of residential connections is in support, and in the cost of the
physical connection, whether it's used or not. From the ISP's
perspective, even if the average customer's use were to drop considerably,
it probably wouldn't lead to a huge reduction in their costs, so they
wouldn't be able to lower the base price of an unused circuit much below
what it already is. While it might be nice to be able to get more than
they're currently getting from customers who are heavy users, the heavy
users would be unlikely to pay more, given that they could get service for
the same flat rate from the ISP's competitors.

From the end users' perspective, we've got pretty much the same story.

They're unlikely to save more than a few dollars if they don't use the
connection at all, and they'll have to pay more if they do. What's in it
for them? If the end user is already paying the $30-50/month you suggest
that they would pay for the loop, then they're currently getting the
bandwidth for free. Why would they want to start paying more?

The situation for users of much bigger connections, where we're talking
bills of thousands or tens of thousands of dollars per month, instead of
$30-50, is quite different. Metro ethernet and OC-whatever connections
generally are billed at 95th percentile utilization, which is a form of
pay as you use.

-Steve

12

What's your cost on managing the bandwidth? You're basically creating on-demand frame circuits, and balancing them is tricky (actually, deciding on an oversubscription ratio is easy, dealing with the customers is the tricky part!) on a low-margin basis. Of course, if you're a BofH or a sales guy, I expect that to be less bothersome than if you're a techie who has to actually talk to the customer when their neighbor takes up their bandwidth. Sometimes I wish I could be a bit more slimey to make paying the bills less painful *sigh*

Something I'd be more interested in for personal use would be protected the usefulness of my site, as well as the cost of it, against a slashdotting. If I get slashdotted on the first of the month, I essentially pay the same as if my site gets slashdotted on the 30th. The difference is in 29 days of downtime. And no, I don't have a solution to offer for that one, but that's what always annoys me - to see a site get slashdotted at the beginning of the month, knowing my attention span won't last until next month :slight_smile:

Rob Nelson
ronelson@vt.edu

13

Steve,

As for your point of the major cost for an ISP would be support. That is where I beg to differ, in my own experience working for this company on this project, it has required very little time to do actual support work to the end-user, provided that the Internet connection actually works.

Steve Gibbard wrote:

For an idea to catch on, it often helps for there to be a clear benefit to
doing things the new way rather than the old way (or at least, it needs
some good marketing...).

The benefit would be better bandwidth management for the ISP. For example, in our building where we're lit, we have 50 apartments currently running this exact type of scenario on an Ethernet platform to a T1 going into the building.

In this case, it's not clear to me where the benefit is. A lot of the
cost of residential connections is in support, and in the cost of the
physical connection, whether it's used or not. From the ISP's
perspective, even if the average customer's use were to drop considerably,
it probably wouldn't lead to a huge reduction in their costs, so they
wouldn't be able to lower the base price of an unused circuit much below
what it already is. While it might be nice to be able to get more than
they're currently getting from customers who are heavy users, the heavy
users would be unlikely to pay more, given that they could get service for
the same flat rate from the ISP's competitors.

As for your point of the major cost for an ISP would be support. That is where I beg to differ, in my own experience working for this company on this project, it has required very little time to do actual support work to the end-user, provided that the Internet connection actually works.

But, to a heavy user, a faster connection (greater than 1.5Mbps downstream, with a higher upstream than most DSL companies provide in their residential packages)would also be worth paying for.

For example, would an office with 100 employees rather work on a single T1 line for which they are paying $1000/month whether they use it or not. Or, would they want to buy bandwidth for the 8-9 hours that they are actually there during the week. Currently, as I type this, the actual bandwidth market in the building is sitting (and has been sitting for quite awhile) at $1.22 per kbps sustained for a month. So, if you don't use it the whole entire month (as your billed in 5 minute increments of actual usage), then your cost becomes significantly less.

With our model, you wouldn't be paying for what you don't use. This has been the way that we lit the apartments in the building as well as how we operate our colocation market.

From the end users' perspective, we've got pretty much the same story.

They're unlikely to save more than a few dollars if they don't use the
connection at all, and they'll have to pay more if they do. What's in it
for them? If the end user is already paying the $30-50/month you suggest
that they would pay for the loop, then they're currently getting the
bandwidth for free. Why would they want to start paying more?

I'm saying bandwidth at market rate + a constant fee for the T1 loop.

The situation for users of much bigger connections, where we're talking
bills of thousands or tens of thousands of dollars per month, instead of
$30-50, is quite different. Metro ethernet and OC-whatever connections
generally are billed at 95th percentile utilization, which is a form of
pay as you use.

Yes, but as I wrote before, this is not a 95th percentile environment, it's actual usage. We are actually AGAINST 95th percentile. It seems to work quite well for the building that we've already wired with this technology.

14

James,

I'd rather keep paying more for unmetered service rather than pay by the
byte. I can host a popular site for a couple months, download a few cds,
upgrade all of my machines, without having to worry about explaining to
my wife why my monthly bill has doubled or tripled.

For $850 a month, I have an unmetered T1. Though I usually only use
15-20 gigabytes a month, on some months I can quadruple that.

But I'm talking about using T3 connections to serve a building or an area.

Basically, with metered service, first I'd become paranoid about usage,
then increasingly unhappy as time passed. I'd be very unsatisfied with
myself as I turned away projects like gnuarch because I didn't know how
much it would cost me to host them. Then, I'd become infuriated at you
for charging me for bandwidth that I didn't want in the first place --
spams, web based ads, so forth and so on.

As for turning away projects - you wouldn't have to. You can set a limit of how much you want to pay and buy bandwidth up to that limit. Also, the market fluctuates as the supply/demand goes up and down.

As for spam, etc., nobody said that you had to download it from your ISP's mailserver. However, I do see your point. To be perfectly honest with you, thats not one of the concerns that our current customers have.

I could just imagine saying to myself "I just downloaded the new Debian
CD. My bastard of a provider just charged me five bucks"

That may be true - but also think about all the time that your *not* paying for the connection while your at work or out doing something else.

Though I pay a lot now, I know that whether I'm using 20 gigs a month,
80, or 400, the bill will remain the same predictable amount.

I agree with you that knowing what your bill is, is very nice. However, if you had a T3 plugged into your apartment building for mass consumption, that would be nicer, wouldn't it? :slight_smile:

Now, if you want to sell me a T1 with a /28 for $600 a month (not
including local loop) , and then each month offer me a $570 rebate for the bandwidth that I don't use -- well, now we can talk!

If we had facilities where you are, then i'm sure it wouldn't be a problem - but we wouldn't have to give you a rebate, as we wouldn't charge it to you in the first place :slight_smile:

-- Jonathan

15

Rob,

What's your cost on managing the bandwidth? You're basically creating on-demand frame circuits, and balancing them is tricky (actually, deciding on an oversubscription ratio is easy, dealing with the customers is the tricky part!) on a low-margin basis. Of course, if you're a BofH or a sales guy, I expect that to be less bothersome than if you're a techie who has to actually talk to the customer when their neighbor takes up their bandwidth. Sometimes I wish I could be a bit more slimey to make paying the bills less painful *sigh*

Actually, our model doesn't allow for oversubscription as it's a committed (meaning you have the bandwidth that you purchased guaranteed to you), dynamic rate.

16

* jslivko@invisiblehand.net (Jonathan M. Slivko) [Sat 15 May 2004, 01:27 CEST]:

Actually, our model doesn't allow for oversubscription as it's a
committed (meaning you have the bandwidth that you purchased guaranteed
to you), dynamic rate.

Ah, falling into the same trap MAE-East-ATM (and -West-) did.

  -- Niels.

17

George,
We do too - it's just a composite of the 5 minute interval samples taken

18

A former employer of mine does something similar to this with a
metro-ethernet product they provide. Customers are dropped a
FastEthernet port and pay a monthly fee based on their byte count. All
of these customers are served off of a wireless point-to-point
microwave network that supported, at most, 150Mbits/sec. This isn't
generally a problem as the peak traffic on the ring was normally no more
than 20 or 30 Mbits, and much more likely to be about 6Mbits otherwise.

So, who pays when a new Microsoft SQL server worm comes out, and many of
those ports start pumping out 100Mbits of traffic on a Saturday morning?

Also, how many people do you have to call in around the country to start
physically unplugging customer ports to enable your operations staff to
access devices and construct access lists?

I'm not personally fond of business models where capacity planning
requires the use of a crystal ball, or a rabbit, a hat, and a wooden
stick.

-doug

19

Well - you could, to save costs, put a T3 (or multiple T3's) into a specific area that you want to serve and then distribute it from there via Ethernet. This is what we're currently doing with a residential/commercial building.

Ah, so you're only talking about inner city applications. Given I live in an area of single family homes, in a town whose primary industry is agriculture (apple orchards). When I think about serving residential users, I've got a very different set of circumstances in mind.

The colo model doesn't work outside urban areas.

20

Actually, our model doesn't allow for oversubscription as it's a committed (meaning you have the bandwidth that you purchased guaranteed to you), dynamic rate.

That would be impressive, in the residential market. BellSouth is rolling out their 2Meg DSL here and TWCable already supports 3 meg on their residential cable, 4 meg on Commercial accounts. I'd be interested in how you'd compete with them AND not have to oversubscribe.

Rob Nelson
ronelson@vt.edu