Hardware Reference
In-Depth Information
6.9.5
SIMULATION RESULTS
This section shows how the CBS can be efficiently used as a service mechanism for
improving responsiveness of soft aperiodic requests. Its performance has been tested
against that of TBS and DSS, by measuring the mean tardiness experienced by soft
tasks:
E
i,j
=
max
{
0
,f
i,j
−
d
i,j
}
(6.5)
where
f
i,j
is the finishing time of job
J
i,j
.
Such a metric was selected because in many soft real-time applications (e.g., multi-
media) meeting all soft deadlines is either impossible or very inefficient; hence, the
system should be designed to guarantee all the hard tasks and minimize the mean time
that soft tasks execute after their deadlines.
All the simulations presented in this section have been conducted on a hybrid task
set consisting of 5 periodic hard tasks with fixed parameters and 5 soft tasks with
variable execution times and interarrival times. The execution times of the periodic
hard tasks were randomly generated in order to achieve a desired processor utilization
factor
U
hard
. The execution and interarrival times of the soft tasks were uniformly
distributed in order to obtain a mean soft load
U
sof t
=
i
c
i,j
r
i,j
+1
−r
i,j
with
U
sof t
going from 0 to 1
−
U
hard
.
The first experiment compares the mean tardiness experienced by soft tasks when
they are served by a CBS, a TBS, and a DSS. In this test, the utilization factor of
periodic hard tasks was
U
hard
=0
.
5. The simulation results are illustrated in Figure
6.16, which shows that the performance of the DSS is dramatically worse than the one
achieved by the CBS and TBS. The main reason for such different behavior between
DSS and CBS is that while the DSS becomes idle until the next replenishing time (that
occurs at the server's deadline), the CBS remains eligible by increasing its deadline
and replenishing the budget immediately. The TBS does not suffer from this problem;
however,
its correct behavior relies on the exact knowledge of WCETs
, so it cannot be
used for supporting applications with highly variable computation times.
Figures 6.17 illustrates the results of a similar experiment repeated with
U
hard
=0
.
7.
As we can see, TBS slightly outperforms CBS, but does not protect hard tasks from
transient overruns that may occur in the soft activities. Note that since the CBS auto-
matically reclaims any available idle time coming from early completions, for a fair
comparison an explicit reclaiming mechanism has also been added in the simulation
of the TBS, as described by Spuri, Buttazzo, and Sensini [SBS95].
Search WWH ::
Custom Search