FIXED-PRIORITY SERVERS - Hard Real-Time Computing Systems: Predictable Scheduling Algorithms and Applications - page 156

Up = 0.69, U(server) = 24.8%, U(DS) = 23.9%

1

Polling
SS
DS
PE

0.8

0.6

0.4

0.2

0

5 10 15 20 25 30

Average aperiodic load

Figure 5.24 Performance results of PS, DS, PE, and SS.

In the plots shown in Figure 5.24, the average aperiodic response time of each algo-
rithm is presented relative to the response time of background aperiodic service. This
means that a value of 1 . 0 in the graph is equivalent to the average response time of
background service, while an improvement over background service corresponds to a
value less than 1 . 0. The lower the response time curve lies on the graph, the better the
algorithm is for improving aperiodic responsiveness.

As shown from the graphs, DS, PE, and SS provide a substantial reduction in the
average aperiodic response time compared to background and polling service. In par-
ticular, a better performance is achieved with short and frequent requests. This can be
explained by considering that, in most of the cases, short tasks do not use the whole
server capacity and can finish within the current server period. On the other hand, long
tasks protract their completion because they consume the whole server capacity and
have to wait for replenishments.

Note that average response times achieved by SS are slightly higher than those ob-
tained by DS and PE. This is mainly due to the different replenishment rule used by
the algorithms. In DS and PE, the capacity is always replenished at its full value at
the beginning of every server period, while in SS it is replenished T s units of time
after consumption. Thus, on the average, when the capacity is exhausted, waiting for
replenishment in SS is longer than waiting in DS or in PE.