DYNAMIC PRIORITY SERVERS - Hard Real-Time Computing Systems: Predictable Scheduling Algorithms and Applications - page 198

R i

2 C i
U s

C i
U s

Q s

C i
4

C i
3

C i
2

C i

Figure 6.20 Worst-case response time of a CBS as a function of the budget.

From the graph shown in Figure 6.20 it is clear that for a given job with constant exe-
cution time C i , the minimum worst-case response time is C i /U s and can be achieved
when C i is a perfect multiple of Q s . In practice, however, task execution time varies,
inducing response time fluctuations due to the bandwidth enforcement mechanism
achieved through deadline postponements. From Figure 6.20 it is also clear that such
fluctuations would be reduced by making the budget very small compared to the av-
erage execution time, so that the server would approximate the ideal fluid server. Un-
fortunately, however, a small budget (which means a short server period) causes the
job to be split in many small chunks, increasing the runtime overhead. As a conse-
quence, to properly set the server granularity T s , the runtime overhead must be taken
into account in the analysis.

TAKING OVERHEADS INTO ACCOUNT

Whenever the budget is exhausted, the server deadline is postponed, so the served job
can be preempted by other tasks with earliest deadline. If denotes the time needed for
a context switch, then the overhead introduced by the CBS can be taken into account
by subtracting such a time from the server budget.

Hence, Equation (6.8) can be

modified as follows:

C i + C i
Q s −

( T s −

R i = Q s + )

C i + C i
T s U s −

( T s −

= T s U s + ) . (6.9)