Article development led by
A proposal to improve the performance
and availability of streaming video and other
bY aiman eRbaD anD chaRLes “bucK” KRasic
the case for
the internet/WeB architectUre has developed to
the point where it is common for the most popular sites
to operate at a virtually unlimited scale, and many sites
now cater to hundreds of millions of unique users.
Performance and availability are generally essential
to attract and sustain such user-bases. As such, the
network and server infrastructure plays a critical role in
the fierce competition for users. Web pages should load
in tens to a few hundred milliseconds at most. Similarly,
sites strive to maintain multiple nines availability
targets—for example, a site should be available to
users 99.999% of the time over a one-year period.
Such stringent standards, however,
are in practice carefully constructed to
exclude problematic content such as
video streaming and interactive virtual environments. If sites counted vid-eo-streaming rebuffer events against
their availability, their multiple nines
availability would vanish in a poof.
Yet, one could argue that a video rebuffer just at the moment of the winning goal in the World Cup is a form
of unavailability just as severe as, say,
a failure that causes a social-network
user to repeat some steps in composing a post. The goal of this article
is to propose methods for adaptive
transport that are ultimately aimed at
bringing the performance and availability of streaming video and other
time-sensitive media in line with that
of traditional Web content. To do so
we have developed an enhanced transport called Paceline.