DYNAMIC PRIORITY SERVERS - Hard Real-Time Computing Systems: Predictable Scheduling Algorithms and Applications - page 196

Hard task load = 0.6

1600

TBS
CBS

1400

1200

1000

800

600

400

200

0

0 20 40 60 80 100 120 140 160 180 200

Computation times variance

Figure 6.18 Third experiment: CBS against TBS with variable execution times.

The advantage of the CBS over the TBS can be appreciated when WCET i >> c i,j .
In this case, in fact, the TBS can cause an underutilization of the processor, due to
its worst-case assumptions. This fact can be observed in Figure 6.18, which shows
the results of a fourth experiment in which U hard =0 . 6, U sof t =0 . 4, the interarrival
times are fixed, and the execution times of the soft tasks are uniformly distributed with
an increasing variance. As shown in the graph, the CBS performs better than the TBS
when tasks' execution times have a high variance. Additional experiments on the CBS
are presented in the original work by Abeni and Buttazzo [AB98].

6.9.6 DIMENSIONING CBS PARAMETERS

This section presents a statistical study to evaluate the effects of the CBS parameters
( Q s ,T s ) on task response times, and proposes a technique to compute the parameters
that minimize the average response time of the served tasks [BB06].

The worst-case response time R i of a job with computation time C i served by a CBS
with bandwidth U s is a function of the server budget Q s . For the sake of clarity, R i is
first derived by neglecting the overhead, and then modified to take the overhead into
account.