Question on 95th percentile and Over-usage transit pricing

Hello NANOG,

I have a fundamental question regarding 95th percentile pricing. I will
make some prerequisite assumptions to set $/Mbps values before posting
my actual question.

Eg., For 1Gbps commitment, I will pay roughly $3/Mbps. Similarly for
10Gbps, 100Gbps I may pay $2/Mbps and $1/Mbps.

This appears like a sub-linear economy of scale pricing model followed
in transit pricing.

Now if I commit 1 Gbps over a 10Gbps provisioned link, I will pay fixed
monthly fee of $3000 for the 95th peak not exceeding the committed rate
of 1Gbps.

Now if my 95th peak is above the committed rate, say, 2Gbps or 4 Gbps or
8 Gbps, I believe I have to pay: $3000 + [over-usage_bandwidth_charges]

Question: Does this over-usage bandwidth charge a linear cost function
or is it sub-linear like the committed bandwidth pricing? I mean, will
it cost me the same $/Mbps as over-usage charges for all 2Gbps, 4Gbps
and 8Gbps 95th percentile peaks? or is it

Over-usage_charges(2Gbps) > Over-usage_charges(4Gbps) >
Over-usage_charges(8Gbps) ?

I will be grateful for the replies!

With regards

Pradeep Bangera
Research Assistant
Institute IMDEA Networks
Madrid, Spain

This answer is going to suck, but it is the truth. In short, the answer, like so many things, is:

When you sign a contract, the overage can be more, same, or less depending on what you negotiate. Of course, what you can negotiate depends on your leverage. If you you have a lot of traffic, or desirable traffic (e.g. inbound traffic), then you can negotiate favorable terms. If not, well, not.

Typically, 1 Gbps commit is not enough to garner a favorable rate on the overage, so expect to pay at least the same as the commit rate on overage.

If you have a lot more, you can negotiate tiers. E.g. The first 10G is $X/Mbps, but if you hit 20G, you get charged 20000 * $Y (where Y < X, obviously). This can lead to interesting situations where 19 Gbps costs more than 20 Gbps. But dems da breaks.

I knew of a place that used to push "fake" traffic over a link to ensure
they were in the cheaper (higher) tier. Who knew business rules overriding
engineering could result in non-optimal situations.

An optimal solution would be a tiered system where the adjusted price only
applies to traffic units over the price tier threshold and not retroactively
to all traffic units.

Optimal for whom?

Also, I doubt you can make that claim as you do not know the costs or other business conditions of every deal.

I have seen a more "optimal" scheme about 15 years ago. Pricing was a
smooth function, but it was for software licensing, not networking.

As I recall, their scheme went something like:
invoice_amount = some_constant * (quantity)^0.75

This seemed smart to me. It gave the customer incentives to invest
more, but also got rid of silly discontinuities that would cause
irrational customer and salesperson behavior.

Has anyone seen something similar in the service provider world? All I
ever see are arbitrary step functions.

I like thisone!

I actually had this discussion quite recently with The Powers, as we have
some fairly interesting issues with the results of our newly adjusted
pricing steps.

The rationale behind sticking with the steps was "everyone else does it that
way, so when customers are making comparisons they need to be able to make a
meaningful comparison" and "continuous functions are too hard". Given that
we're not a market leader in network traffic, I somewhat see the logic
behind the first, and given the average customer has trouble understanding
that XGB per month at $Y/GB => $X*Y, I totally see the point on the second,
*in general*.

However, if you want it, ask for it. Go so far as to say that you'll only
consider pricing functions that are continuous, and therefore will be making
an apples-for-apples comparison. You'll exclude a lot of the market, simply
because the contracts can't be modified like that or the billing system
can't handle it, but I'm fairly confident that the data to create such a
function exists at every sanely-run network provider.

- Matt

I don't know every particular deal, but I felt it's a solution to the
person's situation whom I replied to who was producing fake traffic for
bandwidth they purchase.

The point is to suggest that his pricing scheme where it's potential for the
total bill to be cheaper by purposely wasting a resource and directing a
traffic flood at the ISP's router or some poor sap's netblock is a
suboptimal one and not in anyone's best interest.

I don't know the costs of his deal, but I think that an arrangement which
has progressed to the customer running a traffic flood to the SP core to get
the bill down is not economically nor politically beneficial to anyone
involved in that person's scenario anymore.

The business conditions can throw a wrench into things though... a huge

* Pradeep Bangera:

Question: Does this over-usage bandwidth charge a linear cost function
or is it sub-linear like the committed bandwidth pricing?

Percentile-based pricing is never linear. It's not even a continuous
function of bandwidth usage. This is inherent to the percentile
functional, so it doesn't matter how the quantity that comes out of that
is priced.