Management of Internet Data Streaming
Environments
Under Peak Rate Pricing
Otis Jennings
Visiting Faculty, Martin Luther King Fellow
MIT Sloan School of Management
Internet firms whose primary
business involves data streaming-whether music, video, or some other
"large-file" medium-often face the problem of reserving bandwidth
from third party Internet service providers under a peak rate pricing model.
The firm reserves an amount of bandwidth at a given price, the peak (often the
95th fractile) bandwidth usage over a time period is observed, and the firm is
charged a higher price for the amount by which the peak usage exceeded the
amount reserved. We show that the optimal bandwidth to reserve is the solution
to a standard news vendor problem. We track the number of concurrent
users/connections to a given site over time, a process that serves as a
surrogate for the bandwidth usage process. Our main results are asymptotic ones
that lead to different approximations of the peak rate of this process. We
examine the sensitivity of our results to changes in the arrival rate of
customers and the distribution of connection times. We show in the latter
context the counter-intuitive result that increased variability in connection
times can improve performance.