1. Optimization of Time-Partitions for Mixed-Criticality Real-Time Distributed Embedded Systems Domițian Tămaș-Selicean and Paul Pop Technical University of Denmark

2. 2 Outline Motivation System and application models Problem formulation and example Optimization strategy Experimental results Conclusions and future work

3. 3 Motivation Real time applications implemented using distributed systems PE1PE2 PE3 Application

4. 4 Motivation Real time applications implemented using distributed systems Mixed criticality applications share the same architecture PE1PE2 PE3 Safety-critical application Non-critical application Mission-critical application

5. 5 Motivation Real time applications implemented using distributed systems Mixed criticality applications share the same architecture Solution: partitioned architectures such as IMA PE1PE2 PE3 PE1PE2 PE3 Safety-critical application Non-critical application Mission-critical application

6. 6 System Model Partition = virtual dedicated machine Partitioned architecture Spatial partitioning Protects one application from another Time partitioning Partitions the CPU time among partitions PE1PE2 PE3 Safety-critical application Non-critical application Mission-critical application

7. 7 System Model : Time Partitions Temporal partitioning Static partition table PE1PE2 PE3 Partition Partition slice PE1 PE2 PE3 Safety-critical application Non-critical application Mission-critical application

8. 8 System Model : Time Partitions Temporal partitioning Static partition table Each partition can have its own scheduling policy PE1PE2 PE3 Partition Partition slice PE1 PE2 PE3 Safety-critical application Non-critical application Mission-critical application

9. 9 System Model : Time Partitions Temporal partitioning Static partition table Each partition can have its own scheduling policy Repeated with a period MF PE1PE2 PE3 Partition Partition slice Major Frame PE1 PE2 PE3 Safety-critical application Non-critical application Mission-critical application

10. 10 System Model : Time Partitions Temporal partitioning Static partition table Each partition can have its own scheduling policy Repeated with a period MF Partition switch overhead PE1PE2 PE3 Partition Partition slice Major Frame PE1 PE2 PE3 Safety-critical application Non-critical application Mission-critical application

11. 11 System Model : Time Partitions Temporal partitioning Static partition table Each partition can have its own scheduling policy Repeated with a period MF Partition switch overhead PE1PE2 PE3 Partition Partition slice Major Frame PE1 PE2 PE3 Safety-critical application Non-critical application Mission-critical application Problem: optimize time partitions

12. 12 Application Model  100120 Priority 88 10 120 60 10 20 CiCi 240 1 2 3 4 100 60 30 120 A2A2    N 2 20 N 2 20 N 1 20    N 1 30 N 1 10 N 2 20 N 2 20 99 1010  11 DiDi TiTi PE A1A1 A3A3 N1N1 N2N2 N1N1 N2N2 D=D=T=T A1A1 = A1A1 A2A2 A2A2 Safety-Critical Applications Static Cyclic Scheduling Non-Critical Applications Fixed Priority Scheduling

13. 13 Problem Formulation Given A set of applications of mixed criticality levels A set of N processing elements (PEs) Mapping of tasks to PEs The size of the Major Frame and of the Application Cycle

14. 14 Problem Formulation Given A set of applications of mixed criticality levels A set of N processing elements (PEs) Mapping of tasks to PEs The size of the Major Frame and of the Application Cycle Determine The allocation of applications to partitions Sequence and length of partition slices for each PE Schedule for the SC applications

15. 15 Problem Formulation Given A set of applications of mixed criticality levels A set of N processing elements (PEs) Mapping of tasks to PEs The size of the Major Frame and of the Application Cycle Determine The allocation of applications to partitions Sequence and length of partition slices for each PE Schedule for the SC applications Such that SC and NC applications are schedulable The available slack is maximized (for upgrades)

16. 16 Motivational Example

17. 17 Motivational Example MF 11 N1N1 A1A1 04080 120 N2N2 22 55 66 A2A2 A3A3 A1A1 A2A2 A3A3 88 88 1010 99  11 All applications miss their deadlines. Deadlines are missed by  ,             (both jobs)      

18. 18 Motivational Example 11 N1N1 A1A1 N2N2 22 55 66 A2A2 A3A3 A1A1 A2A2 A3A3 88 88 1010 99  11 All applications miss their deadlines. Deadlines are missed by  ,             (both jobs)       11 N1N1 A1A1 04080 120 N2N2 22 77 66 33 44 A2A2 A1A1 A1A1 A2A2 A3A3 55 A2A2 88 88 1010 99 A3A3 6020100  11 SC applications meet their deadlines. NC application miss theirs. Deadlines are missed by:  8,  9 and  10

19. 19 Motivational Example 11 N1N1 A1A1 N2N2 22 77 66 33 44 A2A2 A1A1 A1A1 A2A2 A3A3 55 A2A2 88 88 1010 99 A3A3  11 11 N1N1 A1A1 04080 120 N2N2 22 77 66 33 44 A2A2 A1A1 A1A1 A2A2 A3A3 55 A2A2 88 88 1010 99 A3A3 6020100 A3A3 A3A3  11 Deadlines are missed only by the NC task  9 on N 2. All other tasks meet their deadlines SC applications meet their deadlines. NC application miss theirs. Deadlines are missed by:  8,  9 and  10

20. 20 Motivational Example 11 N1N1 A1A1 N2N2 22 77 66 33 44 A2A2 A1A1 A1A1 A2A2 A3A3 55 A2A2 88 88 1010 99 A3A3 A3A3 A3A3  11 Deadlines are missed only by the NC task  9 on N 2. All other tasks meet their deadlines 11 N1N1 A3A3 04080 120 N2N2 22 77 66 33 44 A2A2 A1A1 A2A2 A3A3 55 A2A2 88 88 1010 99 6020100 A1A1 A3A3 66 A2A2 A3A3  11 Deadlines are met for the all the applications A1A1

21. 21 Optimization Strategy Time Partition Optimization (TPO) strategy: Simulated Annealing meta-heuristic The allocation of applications to partitions Sequence and length of partition slices for each PE List scheduling Schedule for the SC applications

22. 22 Optimization Strategy Time Partition Optimization (TPO) strategy: Simulated Annealing meta-heuristic The allocation of applications to partitions Sequence and length of partition slices for each PE List scheduling Schedule for the SC applications Simulated Annealing Minimizes the cost function Explores the solution space using design transformations

23. 23 Optimization Strategy: Cost Function Degree of schedulability Captures the difference between the worst-case response time and the deadline The response time for SC is obtained through List Scheduling The response time for NC is obtained using a Response Time Analysis

24. 24 Optimization Strategy: Cost Function Degree of schedulability Captures the difference between the worst-case response time and the deadline The response time for SC is obtained through List Scheduling The response time for NC is obtained using a Response Time Analysis Cost Function Weighted combination of the degree of schedulability for SC and NC applications

25. 25 Optimization Strategy: Design Transformations 0235 6 8911 1316 10 1471214 15 x 10 ms

26. 26 Optimization Strategy: Design Transformations 0235 6 8911 1316 10 split 1471214 15 x 10 ms

27. 27 Optimization Strategy: Design Transformations 0235 6 8911 1316 10 split resize 1471214 15 x 10 ms

28. 28 Optimization Strategy: Design Transformations 0235 6 8911 1316 10 split resize swap 1471214 15 x 10 ms

29. 29 Optimization Strategy: Design Transformations 0235 6 8911 1316 10 split resize swap join 1471214 15 x 10 ms

30. 30 Experimental Results Benchmarks 10 synthetic 2 real life test cases from E3S Evaluated using Straightforward Solution (SS) Allocates to each application a partition size proportional to the utilization of the tasks mapped on that PE Time Partition Optimization (TPO) For 120 minutes w SC = 400 w NC = 100

31. 31 Experimental Results

32. 32 Experimental Results

33. 33 Experimental Results

34. 34 Experimental Results

35. 35 Conclusions Mixed criticality systems, with safety-critical and non-critical applications running on the same processor, are implemented using a partitioned architecture. The safety-critical applications are scheduled using Static Cyclic Scheduling. The non-critical applications are scheduled using Fixed Priority Scheduling. We proposed a SA based optimization of time partitions. Optimization of time partitions is needed for mixed criticality applications implemented on partitioned systems.

36. 36 Future work Map tasks to PEs Consider other meta-heuristics, Tabu Search Allow partition slice sharing among applications Consider certification costs

37. 37 Thank you!

38. 38 Experimental Results