1. EE 249, Fall 20021 Discussion: Scheduling Haibo Zeng Amit Mahajan

2. EE 249, Fall 20022 Outline Problem: multi-program scheduling on a single processor Optimum fixed priority scheduler – rate monotonic scheduling Optimum dynamic scheduling algorithm – deadline driven scheduling Mixed scheduling algorithm

3. EE 249, Fall 20023 Introduction Environment: hard-real-time vs. soft-real-time Scheduling: Preemptive and priority driven Fixed priority vs. dynamic

4. EE 249, Fall 20024 Assumption of environment A1: The requests for all tasks with hard deadlines are periodic A2: Task deadline is its next request A3: The tasks are independent A4: Run-time for each task is constant A5: Nonperiodic tasks have no deadlines

5. EE 249, Fall 20025 Outline Problem: multi-program scheduling on a single processor Optimum fixed priority scheduler – rate monotonic scheduling Optimum dynamic scheduling algorithm – deadline driven scheduling Mixed scheduling algorithm

6. EE 249, Fall 20026 Rate Monotonic Scheduling According to their request rates only Higher request rates, higher priorities Optimum fixed priority scheduling: feasible if any feasible fixed priority assignment exists Proof: Use the concept critical instant to analyze the case of scheduling two tasks

7. EE 249, Fall 20027 RMS (cont’d) Proof: Use the concept critical instant to analyze the case of scheduling two tasks Result: assign higher priorities to task with shorter request period; independent of their run-times. Generalize this result to m tasks

8. EE 249, Fall 20028 RMS (cont’d) Available Processor Utilization can be as low as: Analysis: Right hand side of the inequality is monotonic decreasing with m

9. EE 249, Fall 20029 RMS (cont’d) Question: Why utilization factor can’t reach 100%? Answer: Processor idle time: example Question: How to relax the utilization bound? Answer: For i=1,2,…,m-1, Better choice: dynamic priority assignment

10. EE 249, Fall 200210 Outline Problem: multi-program scheduling on a single processor Optimum fixed priority scheduler – rate monotonic scheduling Optimum dynamic scheduling algorithm deadline driven scheduling Mixed scheduling algorithm

11. EE 249, Fall 200211 Deadline Driven Scheduling According to their request rates: earliest deadline first (EDF) No processor idle time before overflow Schedulable iff processor use is less than 1 Optimum scheduling algorithm: feasible if any feasible assignment exists

12. EE 249, Fall 200212 Outline Problem: multi-program scheduling on a single processor Optimum fixed priority scheduler – rate monotonic scheduling Optimum dynamic scheduling algorithm – deadline driven scheduling Mixed scheduling algorithm

13. EE 249, Fall 200213 Mixed scheduling algorithm Nice for many applications Interrupt hardware: fixed priority scheduler Other software tasks: dynamic priority scheduler Scheduling algorithm K tasks of shortest periods: RMS Remaining slower paced tasks: EDF

14. EE 249, Fall 200214 Mixed scheduling algorithm (cont’d) Comparison with RMS and EDF: Still can’t reach 100% utilization But much better than RMS Example with 3 tasks T 1 =3, T 2 =4, T 3 =5 C 1 =1, C 2 =1, C 3 =1(rate-monotonic), 2(mixed) RMS: U = 1/3 + 1/4 + 1/5 = 78.3% Mixed scheduling algorithm: U = 1/3 + 1/4 + 2/5 = 98.3%

15. EE 249, Fall 200215 Questions Overhead we ignored? Dynamic scheduling Preemption If programs are nonterminating, how about the resources? Are the assumptions about environment always fine? If A1 or A4 don’t hold, what do we do? Is A3 suitable for embedded systems?