The goal of this more mathematically demanding chapter is to explain some recent insight into the
operations of network protocols. Two points are noteworthy. First, TCP is an approximation of a
distributed algorithm that maximizes the total utility of the active connections. Second, a new class
of backpressure protocols optimize the scheduling, routing, and congestion control in a uniﬁed way.
8.1 THE ROLE OF LAYERS
The protocols regulate the delivery of information in a network by controlling errors, congestion,
routing, and the sharing of transmission channels. As we learned in Chapter 2, the original view was
that these functions are performed by different layers of the network:layer 4 for error and congestion
control, 3 for routing, and 2 for the medium access control. This layer decomposition was justiﬁed
by the structure that it imposes on the design and, presumably, because it simpliﬁes the solution of
However, some suspicion exists that the ‘forced’ decomposition might result in a loss of
efﬁciency. More worrisome is the possibility that the protocols in the different layers might interact
in a less than constructive way.To address this interaction of layers, some researchers even proposed
some ‘cross-layer’ designs that sometime resembled strange Rube Goldberg contraptions.
Recently, a new understanding of control mechanisms in networks emerged through a series
of remarkable papers. Before that work, schemes like TCP or CSMA/CA seemed to be clever but
certainlyad hoc rules for controlling packettransmissions:slowdownif thenetwork seems congested.
For many years, nobody really thought that such distributed control rules were approaching optimal
designs. The control community by and large did not spend much time exploring TCP.
The new understanding starts by formulating a global objective: maximize the total utility
of the ﬂows in the network. The next step is to analyze the problem and show that, remarkably, it
decomposes into simpler problems that can be thought of as different protocols such as congestion
control and routing. One might argue that there is little point in such after-the-fact analysis. How-
ever, the results of the analysis yield some surprises: improved protocols that increase the utility of
the network. Of course, it may be a bit late to propose a new TCP or a new BGP. However, the
new protocols might ﬁnally produce multi-hop wireless networks that work. Over the last decade,
researchers have been developing multiple heuristics for routing, scheduling, and congestion control
in multi-hop wireless networks.They have been frustrated by the poor performance of those proto-
cols. A common statement in the research community is that “three hops = zero throughput." The
protocols derived from the theoretical analysis hold the promise of breaking this logjam. Moreover,