Performance Evaluation
5.2 Experimental Results
5.2.3 The Effect of the Performance on the Number N p of the Periodic Tasks
The figures in this subsection are the experimental results under different Np when Ue = 1.3 and Up = 0.7.
The Effect of System Energy Consumption on Np When W CETBCET = 0.1 Figure 5.11 shows the normalized energy consumptions under the RRA scheme with different user-defined ratio values and the MRA scheme. The normalized total energy consumption is defined as the ratio of the total energy consumption derived by the proposed RRA scheme divided that derived by the MRA scheme, i.e., RRA100%, when the task number is equal to 5.
We can observe that under different periodic task numbers, the total system energy con-sumption when R = 10% is larger than that of the other ratio settings, i.e., R = 50%, R = 90%, and R = 100% (i.e., the MRA scheme). This is because in our proposed RRA scheme with a small ration R value, the periodic jobs can reclaim a low specific ratio of ear-liness to reduce speed. Hence, there are potentially more left earear-liness for the execution of the aperiodic jobs in the future. When Ue = 1.3 and Up = 0.7, then Ue = 0.6. Due to
BCET
W CET = 0.1, the actual aperiodic workload is light which is about 0.06 (= 0.6∗ 0.1). Then,
Energ y*Av rgRe sp T Ime
Figure 5.7: Energy consumption*Average response Time with varying Upwhen W CETBCET = 0.1
Energ y Con sump tion
Figure 5.8: Energy consumption with varying Up when W CETBCET = 0.5
the earliness is only used to execute aperiodic jobs early and thus, the more unused earliness is re-put into the earliness-queue. Little earliness is used by periodic jobs. However, most earliness will be decreased with time elapsed when the system is idle. Therefore, the smaller R value results in a higher system energy consumption.
We find that under different ratio R settings, when there are more periodic tasks, the total system energy consumption is smaller. The reason is that the more unused earliness in the earliness-queue is used by the periodic tasks to reduce the speed. Besides, the opportunity of the dissipating earliness decreases when the system is idle; in other words, the unused earliness can be employed in reducing the speed of the periodic jobs.
The Effect of the Response Time on Np When W CETBCET = 0.1. Figure 5.12 shows the normalized average response time under the RRA scheme with different user-defined ratio values and the MRA scheme. The normalized average response time of the aperiodic job is defined as the ratio of the average response time of the aperiodic jobs derived by the proposed RRA scheme with different R values divided that derived by the MRA scheme, i.e., RRA100%
when the task number is equal to 5.
We observe that different task numbers do not affect the response time of the aperiodic jobs. That is because the proposed RRA scheme assigns all of the earliness in the
earliness-Avera ge R espo nse T ime
Figure 5.9: Average response time with varying Upwhen W CETBCET = 0.5
Energ y*Av rgRes p Tim e
Figure 5.10: Energy consumption*Average response time with varying Upwhen W CETBCET = 0.5 queue to the aperiodic jobs while they execute without considering whether periodic jobs are in the ready queue or not. Besides, we find that under different periodic task numbers the average response time when R = 10% is lower than that when R is set as different values, i.e., 50%, 90%, and 100%, (i.e., the MRA scheme). The reason is that when R = 10%, just a small ratio of the earliness is used to reduce the speed of the periodic jobs. The remaining earliness is utilized to improve the average response time of the aperiodic jobs.
The Effect of the System Energy Consumption and the Average Response Time on Np
When W CETBCET = 0.1. Figure 5.13 shows the system performance, the product of the system energy consumption and the average response time of the aperiodic jobs under the proposed RRA scheme with different ratio values and the MRA scheme. We observe that the perfor-mance under RRA50% is better than the other ones. The reason is that when R = 50%, the proposed RRA scheme can reclaim more earliness for periodic jobs to reduce energy con-sumption and cover the slightly raised response time as shown in Figure 5.11 and 5.12. Thus, the system performance of RRA50% is the best one.
The Effect of System Energy Consumption on Np When W CETBCET = 0.5. Furthermore, Figure 5.14 shows the normalized energy consumption under the RRA scheme and the MRA scheme when W CETBCET = 0.5. We find that the trends in Figures 5.11 and 5.14 are similar
but in the later one the compared schemes have higher system energy consumption. That is because compared with W CETBCET = 0.1 when under W CETBCET = 0.5 the system has a higher actual workload.
The Effect of the Response Time on Np When W CETBCET = 0.5. Figure 5.15 shows the normalized average response time under the RRA scheme and the MRA scheme. The normal-ized average response time is defined as the ratio of the average response time of the aperiodic jobs derived by the proposed RRA scheme divided that derived by the MRA scheme when the task number is equal to 5 and W CETBCET = 0.1. We find that the average response time when R = 100% is the largest. This is because all the earliness is used to reduce the speed of the periodic jobs. Thus, it results in that the aperiodic jobs are delayed critically.
The Effect of the System Energy Consumption and the Average Response Time on Np When W CETBCET = 0.5. In Figure 5.16, we show the system performance under the proposed RRA scheme with different ratio values and the MRA scheme. We find that the trends of the system performance are similar to that when W CETBCET = 0.1. The RRA scheme with a high ratio value, i.e., RRA90%, and RRA100%(i.e., the MRA scheme), has the worst performance.
That is because more earliness is reclaimed by periodic jobs and then the response time of the aperiodic jobs becomes larger.
Energ y Con sump tion
Figure 5.11: Energy consumption with varying Npwhen W CETBCET = 0.1
Based on our experimental results, we observe that when the actual system workload ranges between 0.13 (= 1.3 ∗ 0.1) to 0.91 (= 1.3 ∗ 0.7), RRA50% or RRA10% is a more reasonable decision. That is because they reclaim more earliness to aperiodic jobs to reduce the average response time and cover the merely raised system energy consumption. Hence, we propose the following rule for the choice of the user-defined ratio R: Whenever the actual system workload is not over-loaded, we prefer a lower R value, i.e., RRA50% or RRA10% to schedule for the mixed task sets.
Avera ge R espon se Ti me
Figure 5.12: Average response time with varying Npwhen W CETBCET = 0.1
Energ y*Av rgRes p TIm e
Figure 5.13: Energy consumption*Average response time with varying Npwhen W CETBCET = 0.1