Hardware Reference
In-Depth Information
An interesting property of EDF during permanent overloads is that it automatically
performs a period rescaling, so tasks start behaving as they were executing at a lower
rate. This property has been proved by Cervin in his PhD dissertation [Cer03] and it
is formally stated in the following theorem.
Theorem 4.7 (Cervin)
Assume a set of
n
periodic tasks, where each task is described
by a fixed period
T
i
, a fixed execution time
C
i
, a relative deadline
D
i
, and a release
offset
Φ
i
.If
U>
1
and tasks are scheduled by EDF, then, in stationarity, the average
period
T
i
of each task
τ
i
is given by
T
i
=
T
i
U
.
Note that under fixed priority scheduling, a permanent overload condition causes a
complete blocking of the lower priority tasks.
As discussed later in the topic, another major advantage of dynamic scheduling with
respect to fixed priority scheduling is a better responsiveness in handling aperiodic
tasks. This property comes from the higher processor utilization bound of EDF. In fact,
the lower schedulability bound of RM limits the maximum utilization (
U
s
=
C
s
/T
s
)
that can be assigned to a server for guaranteeing the feasibility of the periodic task set.
As a consequence, the spare processor utilization that cannot be assigned to the server
is wasted as a background execution. This problem does not occur under EDF, where,
if
U
p
is the processor utilization of the periodic tasks, the full remaining fraction 1
−
U
p
can always be allocated to the server for aperiodic execution.
Exercises
4.1
Verify the schedulability and construct the schedule according to the RM algo-
rithm for the following set of periodic tasks:
C
i
T
i
τ
1
2
6
τ
2
2
8
τ
3
2
12
Search WWH ::
Custom Search