44
Degree of
realism
Experimental
perspective
Network
Application
Simulation Emulation Direct
Measurement
Analytic
Modeling
Martin
Holt
Hall
Chang
Anh
Chiu
Huang
Wilson
Figure 2.8:
Methodologies in Context
This figure shows were various works fit in the space of network experiment methodologies. The x-
axis orders methodologies by degrees of realism, while the y-axis shows the focus of the dependent
variable.
2.6.5
Summary
In this section, we return to the network experiment conceptual framework developed in
Section 1.1. Figure 2.8 shows a number of studies placed in our framework of experiment design.
The experiments are: Holt [52], Martin [73], Hall [49], Chang [21], Ahn [1], Chiu [24], Huang [54]
and Wilson [62, 66, 67]. When placing experiments in this space, we have used the primary focus
of the study to determine the axis. Although some studies [1, 52, 73] used multiple techniques at
once, it is interesting that even these have one axis clearly dominate the other.
We can see that experiments cover nearly the full spectrum of experiment design. A point
relevant to this thesis is that there are few experiments which categorize application-centric sensitiv-
ities to abstract parameters. Thus, there is little systematic exploration of the system design space.
Instead, most studies compare single instances of different systems or algorithms, e.g. TCP Vegas
vs. Reno [1], credit vs. rate based flow control [21], or congestion avoidance algorithms [24]. A
point which does not need elaboration is that, of course, many research groups have an axe to grind
with regard to these systems.
There is a notable lack of application-centric experiments which rely on analytic model-
ing. Although [24], and to a lesser extent [56] model the effects on applications somewhat, the ac-
tual concern is not applications per say, but the collective effects of many applications on the net-
work infrastructure. Speculating a bit, the author believes this is because applications are difficult