54 / 64
5 4
p ro j e ct 1 4
Most of the practical network protocols were
designed based on sound yet ad-hoc heuristics.
They have performed reasonably well as the
Internet scaled up by six orders of magnitude.
With the Internet continuing to grow, current
protocols need to be re-engineered and
optimized for emerging mobile applications.
Both engineering heuristics and optimization-
theoretic approaches were recently employed
to enhance/re-engineer the existing network
protocols. The resultant schemes, however,
either cannot provide analytical performance
guarantees, or are difficult to be deployed
with the current Internet infrastructure. To
Design-Space Oriented Cross-Layer Optimization for Mobile
Internet Applications
Xin Wang, PI
fill the gap, this project aims to developing
a systematic approach to design and analysis
of readily deployable, scalable, yet optimal
network schemes for mobile Internet
applications. As core of the Internet protocols,
the TCP performs window-based congestion
control for competing data flows. Adopting
queueing delay as the congestion measure, we
reveal that TCP window-control mechanism
amounts to implicit updates of source rates
as “primal variables” and queueing delays as
“Lagrange dual variables” from optimization-
theoretic perspective. Confining to the design
space of Internet, we then envision that the
keys to cross-layer Internet optimization are
development of non-standard window-control
oriented implicit primal-dual solvers for
underlying utility maximization problems, and
design of jointly optimal network protocols as
decomposition of such solvers.
Capitalizing on this original idea, we consider
an Internet with wired backbone and a
single access point that provides one-hop
communications for mobile devices to access
Internet. The proposed research will bridge the
gap between the current network optimization
theory and practical Internet designs, and is
expected to benefit directly applications to
next-generation Internet protocols and designs.