| |
|
Previously, I mentioned
that InterStream must play some role establishing the standards for the
quality and objective price comparison of media grid services. As we
move through our pilot, we
have to keep our overall intentions to create a broadly available new
bandwidth commodity known as media grid service. In order to gain an
understanding of both where we need to go with the service as well as
how to price it, we need take a quick examination of the past.
IP Transit and the World Wide Web
Originally, the Internet was a network of networks interconnected by
government and educational institutions. As the Internet became
commercialized in the mid-90s, a new paradigm had to be established.
The world wide web implied a global reach. Therefore, a commodity
bandwidth service had to be developed to ensure that reach. That
service, today, is known as IP Transit.
IP Transit is purchased on a monthly basis and paid for in megabits per
second. A similar pricing scheme can be used for media grid services.
However, this time absolute capacity measures don't have to be used.
Content providers, or "streamcasters" in our case, tend to think of
buying bandwidth in terms of the number of bytes streamed rather than
the number of megabits per second per month. Providers of IP Transit
like to sell it that way because they think in terms of offering
network capacity. This approach is confusing to these customers because
they must purchase an absolute amount of capacity continuously that
could be used 24 hours a day, 7 days a week. If the customer uses that
capacity inefficiently, then they are overpaying for the service and
must figure out how to get their end-users to consume that bandwidth
more evenly throughout the day and week.
As I've previously articulated,
we term the efficiency in which customers use the infrastructure as the
peak-to-mean ratio, or P:M. The higher the P:M, the more volatile the
customer's use of that capacity. Lower P:Ms mean the streamcaster is
using that capacity and through aggregation of demand can support their
business models through improved economies of scale. By assigning an
incentive curve which can map the cost in terms of gigabytes streamed
against the peak-to-mean ratios, streaming bandwidth can be priced in
terms of bytes while offering an incentive to those streamcasters who
use the infrastructure more efficiently. (See the graph below.)
Establishing Premium Services with a Global Reach
Currently, IP Transit service reaches globally under a best-effort delivery model.
On the global network, there are primarily two places where there is
constrained capacity or bottlenecks. In order for InterStream to
deliver that same global reach best-effort service offers, those
bottlenecks must be circumvented or mediated.
Between major geographies, like North America and Europe, bandwidth
should get more expensive than within one of those continents. Capacity
is limited across oceans and service cannot be easily guaranteed on
those portions of the Internet's backbone today. Similarly, within the
last-mile, bandwidth is also a precious commodity. InterStream Internet
Service Providers (ISPs) set the prices for their bandwidth for those
last-mile bottlenecks by applying mediation policies to them.
Measuring the Global and Local Reach
Globally, media grid service must be available in any region of the world that InterStreamsm
serves. Therefore, peak-to-mean ratios (P:Ms) must be measurable during
our pilot implementation so that we can determine how to establish the
incentive pricing curve as well as having a good understanding of how
efficiently customers will be using the InterStream bandwidth. In
addition, the "working set" for locally stored content has to be
determined. The geographic working set or a model whereby the global
"tier zero" guarantee service between geographies must be established.
Either way, core statistics must be gathered to determine which
approach is most economically viable for ISPs to use. With the right
understanding of network conditions across multiple network conditions,
content can be stored anywhere, both locally and from a single point
for a more efficient (ie. lower P:M) network infrastructure.
Locally, that is within the last-mile, there are two key sets of
metrics that will be established during the InterStream pilot. These
the mediation metrics and Stream Blocking Probability (SBP).
The mediation metrics specify how bandwidth will be allocated for
different applications. Of course, during our pliot test we will not be
specifying exactly what the thresholds will be for various
applications. However, we will be to establish approximations for these
thresholds across different last-mile connections. See our wiki for more details.
The Stream Blocking Probability (SBP) specifies the likelihood that a
particular ISTP stream will complete its transmission without
interruption. Of course the Association would like to set the quality
standards for this metric as low as possible. By doing so, end-users
will the highest quality of experience possible. However, there are two
different conditions in which this might occur. On some bottlenecks,
which are unmediated, the SBPs could be set at a different level than the mediated ones. The pilot test must clearly establish what the SBPs should and will be for an ISP to claim itself as a provider in the InterStreamsm Association. (See the diagram, below.)
Conclusion
During our pilot test, the following key metrics will be used in managing our test results:
Global:
Peak-to-Mean (P:M)
Pricing Incentive Curve
Geographic Working Set
Local:
Stream Blocking Probability (SBP)
Mediation Policy Metrics
Our InterStream pilot test plan establish media grid service as a new
premium bandwidth commodity on the Internet's backbone by using these
metrics to set pricing policies as well as the quality standards for
the Association. We're committed to maintaining a transparent and open
process for both the pilot test and establishment of our quality
standards. If you represent an ISP, please feel free to contact us
at info <at> interstream.com for more information.
Jeff Turner
|
|
|
