How to game the system (was Re: What does 95th %tile mean?)

How do you game the system? Economists like to write a lot of
papers on this subject.

The problem with calculating the average is the fencepost error.
Warning, majore simplifying assumptions ahead...

If you transfer 1GB for exactly 5 minutes, what is your 95% bill?
It depends on your timing.

  00:00 Start transfer, Start measurement window
  01:00
  02:00
  03:00
  04:00
  05:00 End transfer, End measurement window

Elapsed time 5 minutes, total bytes 1000000, or 3,333 Bps.

Bandwidth billed 3.3Kbps peak

  00:00 Start measurement window #1
  01:00
  02:00
  02:30 Start transfer #1
  03:00
  04:00
  05:00 End measurement window #1 (500,000 Bytes, 300 Seconds)
          Start measurement window #2
  06:00
  07:00
  07:30 End transfer #1 (1,000,000 Bytes, 300 Seconds)
  08:00
  09:00
  01:00 End measurement window #2 (500,000 Bytes, 300 Seconds)

Bandwidth billed 1.6Kbps peak

By knowing your provider's measurement windows, you could cut your
usage bill in half while transfering the same amount of data.

A customer with a relatively random usage (user's surfing the
web) couldn't do this, but a user transfering batch files on a
set schedule may see dramatic differences in their bill.

You can make the measurements more complex by using three windows,
or choosing three out of five windows, and so forth. Unless providers
or users are seeing real billing problems, there is little benefit
to the added complexity.

[ On , April 19, 2001 at 19:12:31 (-0700), Sean Donelan wrote: ]

Subject: How to game the system (was Re: What does 95th %tile mean?)

By knowing your provider's measurement windows, you could cut your
usage bill in half while transfering the same amount of data.

Hmmm, but is that any different than rate limiting your own connection?

If you know the provider's sample window is every five minutes and you
only leave your interface up for 2.5 minutes out of every five, then yes
you'll be able to move data at line rate 50% of the time and only be
billed as if you were running at 50% of line rate. So!?!?!?!? You gain
what, exactly?

From the provider's perspective this doesn't really matter a whole lot

and they can reduce the effect of your bursts on their other customers
by simply narrowing their window. There's no need to do silly sliding
windows, multiple windows, etc. -- the averages are just fine.

A customer with a relatively random usage (user's surfing the
web) couldn't do this, but a user transfering batch files on a
set schedule may see dramatic differences in their bill.

If the user doesn't want to pay for line rate for big batch transfers
then there are a lot better ways to rate-limit their usage in those
cases than to turn their router interfaces on and off at 50% the ISP's
rate sampling window. One way or another though the user has to move N
bytes in T seconds and whether they move it in short bursts or at a
constant average rate half the speed of the bursts is irrelevant.

Either way the ISP doesn't care -- the user pays for the average peak
*bandwidth* over a given sample time that they're actually using!

BTW, with the price of disks these days it's not too difficult for any
ISP to use even a 5-second sample period. That's just over 0.5 million
samples per customer per month, and in a well packed data file that's
just 6MB per customer per month even if you keep a timestamp and both
the in and out Counter32 values! Spread the data out a bit with field
separators and you're still only looking at less than 7MB/customer/month
in the worst of cases. Tell me again how the customer's going to beat
the system and get more value than I bill him for? Not only that but
it's now enough of a bother that the customer probably won't bother
trying to collect his own samples and match my calculations! :slight_smile:

There is a much simpler game that costs the ISP a lot more
money. Fortunately, it's not a common business model.

  Let's say I am a TV network, and I want to simulcast a TV
show once a week to the Internet. I might need 2-3 Gig of capacity
during the simulcast, but the rest of the time I need none. So,
I buy 95% service, stream for 4 hours a month, which is thrown away
in any of the counting schemes put forth so far, and pay nothing.

  Lather, rinse, repeat with each TV show. There's no incentive
to buy a bundle of service and stream all the shows (more approximating
continuous usage) from one place.

  Fortunately this application is small, but if you were a
web hoster you could do the same thing with multiple providers.
With 20 providers, you could move your bandwidth with that provider
only 5% of the time, paying nothing for service with any of them.

The one that surprises me is that carriers haven't introduced the concept
of "off-peak" billing. Under my 95% billing from upstreams at shore.net,
we could have packed the off-peak hours with data and not impacted our
costs at all. It doesn't seem too difficult to shop for enterprise
customers who need to move batch data from one place to another, sell them
a circuit, and give them a sharply reduced rate as long as their peaks
happen between 12:00 am EST and 6:00 am EST. All I'm doing at that point
is filling the "valleys" on my transit graphs with free data, since I'm
paying for (almost) peak utilization anyway.

-travis

One of the ways many providers accidentally protect against this is by
requiring a minimum commitment of 10 Mbps on 100 Mbps circuits and 100
Mbps on 1000 Mbps circuits. By requiring 10% utilization they are still
statistically likely to make money on people who attempt to game the
system with a 5% duty cycle. In large numbers, 100%/5% = 20 gamers/pipe,
20 x 10% = 200% revenue commitment for said pipe with 100% utilization.

  There is a much simpler game that costs the ISP a lot more
money. Fortunately, it's not a common business model.

  Let's say I am a TV network, and I want to simulcast a TV
show once a week to the Internet. I might need 2-3 Gig of capacity
during the simulcast, but the rest of the time I need none. So,
I buy 95% service, stream for 4 hours a month, which is thrown away
in any of the counting schemes put forth so far, and pay nothing.

  Lather, rinse, repeat with each TV show. There's no incentive
to buy a bundle of service and stream all the shows (more approximating
continuous usage) from one place.

  Fortunately this application is small, but if you were a
web hoster you could do the same thing with multiple providers.
With 20 providers, you could move your bandwidth with that provider
only 5% of the time, paying nothing for service with any of them.

--
Leo Bicknell - bicknell@ufp.org
Systems Engineer - Internetworking Engineer - CCIE 3440
Read TMBG List - tmbg-list-request@tmbg.org, www.tmbg.org

+------------------- H U R R I C A N E - E L E C T R I C -------------------+

[ On Friday, April 20, 2001 at 12:56:54 (-0400), Travis Pugh wrote: ]

Subject: Re: How to game the system (was Re: What does 95th %tile mean?)

The one that surprises me is that carriers haven't introduced the concept
of "off-peak" billing.

I don't know about the carriers, but some of the littler 2nd and 3rd
tier guys are billing at 95th percentile because that's how they are
billed upstream.

I too would think that if you're actually paying a fixed rate for a
fixed-rate pipe then you'd be far more interested in filling the
valleys, so to speak. I know some broadband providers who are in that
boat, but they don't want to also do what it would take to attract the
traffic that would fill their valleys.