Digital Signal Processing Reference
In-Depth Information
3.4.2.4 Mapping to Objective Space
④
The identified knobs and environment dynamics can be mapped to an objective
space. In this objective space each combination of a certain environment and knob
setting corresponds to one point, representing the obtained performance values for
all objectives. For best-effort systems, absolute mapping is not necessary. Rather, the
actions and environment can be mapped to a marginal objective space which means
that actions correspond to an improvement on a certain dimension rather than an
absolute performance, which is usually easier to define. An example is for instance
a delay constraint: for real-time systems, the delay constraint is a hard constraint,
while for best effort systems, the delay constraint is often relative.
3.4.2.5 Cluster and Monitor System Scenarios
⑤⑥
In traditional scenario-based design, a scenario describes the use of a future system,
in an early phase of the development process. Typically, they appear like narrative
descriptions of envisioned usage episodes, called
use-case
scenarios. As in [19, 47],
we consider a different and complementary type of scenarios, called system scenar-
ios. These scenarios are used to reduce the system cost by using information about
what can happen at RT to make better design decisions at DT and to exploit the
time-varying behavior at RT.
System scenarios are the result of a clustering process in the (marginal) objective
space and, hence, group RT situations, which have similar performance characteris-
tics (as a function of the knob settings). At RT these system scenarios then need to
be able to be monitored and identified. The scenario should be selected that is closest
to the current system state. Clearly, a trade-off exists between adaptation accuracy
and the number of system states that need to be considered. Each additional system
scenario results into memory and monitoring overhead. However, more scenarios
result into better adaptation and, hence, increased gains.
System scenario:
(S
i,m
)
,1
s
. As wireless terminals are operat-
ing in very dynamic environments, the system behavior varies over time
s
. Environ-
mental or application dynamics independent of the user or system's control exist,
represented by the system scenario variable,
S
i,m
.
Monitors:
(M
i
)
,1
≤
i
≤
n
,1
≤
m
≤
m
. For a given wireless environment,
m
monitors can
be defined. A monitor is an explicit observation of an independent parameter of
the observed RT situation. An independent parameter denotes a parameter that the
wireless terminal cannot influence. As a result, monitors are a function of the en-
vironment and the environment only. This is in contrast to the feedback, which we
formally define later in the chapter. Unlike a monitor, the feedback is a function
of the environment and the actions taken by the wireless terminal and allows RT
calibration.
≤
i
≤
