This sub-section describes the assessment conducted by comparing the LBP pro- tocols with the former BP approaches. I have conducted different experiments, each consisting of a group of three thousand implicit deadline task sets randomly generated. The following scheduling protocols have been compared:
• the standard Fixed-Priority Preemptive Scheduling with DM as priority assignment (FPPS-DM).
• the standard Bailout Protocol (BP). • the Lazy Bailout Protocol (LBP).
• the Soft Lazy Bailout Protocol (SLBP).
• the Bailout Protocol - Slack (BPS), Lazy Bailout Protocol - Slack (LBPS) and the Soft Lazy Bailout Protocol - Slack (SLBPS) deriving from the in- tegration of the basic mixed-criticality protocols with the offline sensitivity analysis [106, 44] while guaranteeing the schedulability according to AMC- rtb [69].
• the Bailout Protocol with Gain time (BPG), Lazy Bailout Protocol with Gain time (LBPG), and Soft Lazy Bailout Protocol with Gain time (SLBPG) where each job that finishes before its optimistic time threshold in Normal mode gives its gain time to increase the time budget of next job ready to be scheduled.
• the Bailout Protocol - Slack and Gain time (BPSG), the Lazy Bailout Pro- tocol - Slack and Gain time (LBPSG) and the Soft Lazy Bailout Protocol - Slack and Gain time (SLBPSG) deriving from the integration of both the offline scaling of CLO of HI tasks with sensitivity analysis and the online
gain time collection with the basic scheduling protocols.
Tables 7.1 and 7.2 contain respectively the results about the task set schedu- lability and global job set completion rates of mixed-criticality scheduling meth- ods. Since these experiments are made with set of tasks having deadline equal to periods, all LO jobs deadlines are considered to be hard. I have also collected data within figures to summarise the results of experiments with dual-criticality task sets and show the results in all the three different scenarios. More precisely, Figure 7.1, Figure 7.2 and Figure 7.3 show the averages of task sets and jobs scheduled while Figure 7.4, Figure 7.5 and Figure 7.6 show how the LO jobs scheduled are distributed.
Figure 7.1 and Figure 7.2 summarise the results in cases where there is criti- cality inversion. In these situations, if no HI job completes within its optimistic threshold estimate CLO, then very likely there will be some new incoming higher
priority LO jobs that will interfere with it. Conversely, Figure 7.3 contains in- formation about cases in which all HI jobs have higher priority than LO jobs, i.e., all the critical jobs have smaller deadlines. This basically leads to have no interference between HI and LO jobs and thus no criticality inversion occurrence during the scheduling process.
Looking both at task set and job set schedulabilities results, it is possible to notice that the standard deadline monotonic approach always schedules jobs only according to priorities. In this case, the percentages of HI or LO jobs successfully
scheduled mainly depend only on their priority, with all LO jobs that meet their deadlines in HC-LP scenario, i.e., LO jobs have smaller deadlines than HI jobs, and all HI jobs that always meet their deadlines in HC-HP scenario, i.e., HI jobs have smaller deadlines than LO jobs.
On the other hand, the mixed-criticality protocols always assure that there are no HI jobs missed regardless of job priorities. The experiments confirm what is stated in Chapter 5 with LBP that always successfully schedules more LO jobs than BP since BP schedules no more than 7.07% of task sets with no jobs missed while LBP can schedule till the 52.33% of task sets with no jobs missed. All figures highlight that the amount of jobs scheduled further increases when the offline and online complementary techniques are used. It is worth to notice that the usage of sensitivity analysis and the gain time mechanism always leads to have the same effects when applied both to the standard or to the lazy bailout methods. A noticeable result is that each LBP-based approach allows to complete more LO jobs within their deadlines than the corresponding standard BP-based protocol. In the whole, according to the criteria defined in Chapter 5, LBPSG and SLBPSG are the protocols that increase more the amount of jobs completed within their deadlines. As an example, LBPSG and SLBPSG schedule between the 43.07% and 58.67% of task sets with no jobs missed compared to BPSG for which the percentage of set of tasks with no jobs missed is at maximum 26.87%. As a conclusion, LBP and SLBP always schedule more LO jobs compared with BP while guaranteeing the same level of performances in processing HI jobs. Each protocol can be further refined by exploiting the system slack time identified offline and the online gain time collection to still increase the amount of lower criticality jobs scheduled. With regard to the formal evaluation criteria introduced in Chapter 5, the results show that LBPS and SLBPS always outper- form BPS, LBPG and SLBPG always outperform BPG and LBSG and SLBPSG always outperform BPSG. Finally, the usage of mixed-criticality protocols is rec- ommended in HP-LP and HC-MP scenarios, i.e., when HI jobs could have lower priorities than LO jobs.
Figure 7.4, Figure 7.5 and Figure 7.6 display the distribution of the LO jobs percentages per task set that are completed within their deadlines. Each schedul- ing protocol is represented by a box-and-wisker diagram with the box itself rep- resenting the range in which at least the 50% of results tend to be concentrated. The box also contains the indication of the median and the mathematical aver- age of all the LO jobs scheduled by the related protocol. The results highlight how the LBP/SLBP-based methods always increase the LO jobs success rate, as defined in Chapter 5, compared with the former BP ones.
HC-LP HC-MP HC-HP Method TSSche d TSSche dHI TSSche dLO TSSche d TSSche dHI TSSche dLO TSSche d TSSche dHI TSSche dLO FPPS-DM 83.03 83.03 100.0 66.13 97.20 66.60 68.80 100.0 68.80 BP 7.07 100.0 7.07 2.63 100.0 2.63 4.40 100.0 4.40 BPG 11.90 100.0 11.90 3.10 100.0 3.10 4.53 100.0 4.53 BPS 17.07 100.0 17.07 16.80 100.0 16.80 21.17 100.0 21.17 BPSG 24.17 100.0 24.17 22.13 100.0 22.13 26.87 100.0 26.87 LBP 27.97 100.0 27.97 32.83 100.0 32.83 52.33 100.0 52.33 LBPG 34.23 100.0 34.23 33.53 100.0 33.53 52.70 100.0 52.70 LBPS 36.27 100.0 36.27 40.13 100.0 40.13 55.53 100.0 55.53 LBPSG 43.07 100.0 43.07 44.07 100.0 44.07 58.67 100.0 58.67 SLBP 27.97 100.0 27.97 32.83 100.0 32.83 52.33 100.0 52.33 SLBPG 34.23 100.0 34.23 33.53 100.0 33.53 52.70 100.0 52.70 SLBPS 36.27 100.0 36.27 40.13 100.0 40.13 55.53 100.0 55.53 SLBPSG 43.07 100.0 43.07 44.07 100.0 44.07 58.67 100.0 58.67
Table 7.1: BP and LBP variants: comparison of task set schedulability (%)