Process Scheduling for RTS Dr Hugh Melvin Dept

Process Scheduling for RTS Dr. Hugh Melvin, Dept. of IT, NUI, G 1

RTS Scheduling Approach • RTS typically control multiple parameters concurrently – Eg. Flight Control System • Speed, altitude, inclination etc. . • Options – Use a single process • Simple/limited approach but can be dangerous • No OS required – Use multiple processes • More complex but more robust – Use threads • Light weight relative to processes Dr. Hugh Melvin, Dept. of IT, NUI, G 2

Cyclic Executive Approach • Single Process while(1){ Task 1; Task 2; . . Task n; } • Encapsulate all tasks within single infinite loop Dr. Hugh Melvin, Dept. of IT, NUI, G 3

Cyclic Executive • Ok for very well defined / periodic tasks with bounded execution times • Need to ensure that tasks cannot block and halt all others • May need to slow it down to meet particular task reqds – Eg. Above example will run as fast as processor can handle tasks • Possible Strategy – Run as fast as required by highest freq task – Use lower harmonics for remaining tasks • Eg. Highest freq Task is 100 Hz – Other tasks at 50 Hz, 25 Hz etc • Possible use counters to control sequence – Possible use of timers to sleep • Limited scope for aperiodic tasks eg. Interrupts • Hardware specific very limited portability Dr. Hugh Melvin, Dept. of IT, NUI, G 4

Cyclic Executive • AS Example – Cyclic executive approach – Each AS station handles well defined tasks – Execution time of each task precisely known – Task schedule customised s. t. total task runtime will facilitate/meet response time – Slack time in each cycle built in for interrupt processing • If total task time exceeded repeatedly, timeout fault Dr. Hugh Melvin, Dept. of IT, NUI, G 5

Multiple Process Approach • Each process is much more simple as handling a small subset of overall system • Overall system – More robust and reliable – More scalable • Can run on multiprocessor machine – Modular • Process code more portable – Protected • Each process operates in isolation with dedicated memory less risk of overall system failure • Downside – Requires an OS to schedule multiple processes – More memory reqd for process overhead – Communication between processes has to be explicitly done (advantage!) Dr. Hugh Melvin, Dept. of IT, NUI, G 6

Multiple Process • Generic Process States – Dormant : Process created but not eligible to execute – Ready: Process released and eligible to execute but not doing so • May be preempted • May have been blocked for resources move to ready • May have reached end of timeslice – Executing: Process being executed – Suspended (Blocked): Process waiting for resource to be freed or self-suspended – Terminated: Process finished execution Dr. Hugh Melvin, Dept. of IT, NUI, G 7

Process State Diagram Pre-empt or Timeslice allocated Ready Blocked/ Self suspended Executing Preempted, timeslice up Resource freed Blocked Resource freed Process terminated Schedule Task Dormant Dr. Hugh Melvin, Dept. of IT, NUI, G Terminated 8

Scheduling for RTS • Fundamental OS function • More critical for RTS – Allocating resources and scheduling tasks to ensure that deadlines are met • Schedule is feasible iff all the tasks start after their release time and complete before their deadlines • Scheduling Policy may be determined – Pre-run-time • Schedule created offline • Not unlike Cyclic Executive approach – Run-time • Schedule determined online as tasks arrive – Needs to be done quickly! Dr. Hugh Melvin, Dept. of IT, NUI, G 9

Scheduling for RTS • Static versus Dynamic Priority – Static Priority Scheduling Alg. • Task priority does not change – Rate Monotonic Alg. – Dynamic Priority Scheduling Alg. • Priority can change over time – Earliest Deadline First (EDF) • Preemptive versus non-preemptive – Preemptive Schedule • Task can be preempted by other tasks • Penalty of context switches – Non preemptive • Task runs to completion unless blocked over resource Dr. Hugh Melvin, Dept. of IT, NUI, G 10

Task Characteristics Ti • Precedence Constraints – Other tasks reqd before task Ti can run • Release Time of task Ti = ri • Phase Φi of task Ti = release time of 1 st instance of task Ti • Execution time ei of task Ti = time (worst case) for task to complete • Period pi = interval between consecutive instances of task Ti (presuming periodic) • Absolute Deadline Di = instant by which task must be complete • Relative Deadline di . . Relative to release time ri Dr. Hugh Melvin, Dept. of IT, NUI, G 11

Task Characteristics Ti pi di ei ri Di Φi Dr. Hugh Melvin, Dept. of IT, NUI, G 12

Simplifications • • • All tasks are periodic Relative deadline is equal to period No precedence constraints No task has any non-preemptible sections Cost of preemption is zero Non CPU resources are infinite eg. Memory / I/O • limited use in the real world! Dr. Hugh Melvin, Dept. of IT, NUI, G 13

Rate Monotonic Scheduling • Run time scheduling, Static priority and Preemptive • Priority inversely related to period • Eg. Given task Ti and Tj where pi < pj – Priority of task Ti greater than Tj • Scheduling decision made when – Current task execution complete – New task released • Recall: In real world, RTS tasks with higher freq typically require shortest response times and are the most critical • Recall task Ti utilisation ui = ei / pi – Overall CPU Utilisation U = Dr. Hugh Melvin, Dept. of IT, NUI, G 14

RM Example Task e p u T 1 1 4 0. 25 T 2 2 5 0. 4 T 3 5 20 0. 25 All Tasks released at time 0 Priority T 1 > T 2 > T 3 Overall U = 0. 9 Sequence 1 st inst Task 1 runs to completion 1 st inst Task 2 runs to completion 1 st inst Task 3 runs for 1 unit. . at Eu=4, Task 1 released preempts Task 3 2 nd inst Task 1 runs to completion. . at Eu =5, Task 2 released 2 nd inst Task 2 runs to completion 1 st inst Task 3 runs for 1 unit. . At Eu = 8, Task 1 released preempts Task 3 3 rd inst Task 1 runs to completion 1 st inst Task 3 runs for 1 unit. . At Eu = 10, 3 rd inst of Task 2 released preempts 3. . At Eu = 15, 1 st inst Task 3 completes. . CPU idle Eu 18 -20 At Eu = 20, all 3 tasks released. . Cycle repeats 1 2 2 3 1 3 2 2 1 3 3 2 1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Execution Units Eu Dr. Hugh Melvin, Dept. of IT, NUI, G 15

RM Scheduling • General schedulability test – If overall CPU utilisation U <= n(21/n -1) • where n = no of tasks • RM will definitely produce feasible schedule – However • RM may produce feasible schedule where – U > n(21/n -1) • i. e. Sufficient but not necessary condition • Recall Example: CPU U = 0. 9 but still schedulable – Depends on particular task characteristics • May need to perform further schedulability analysis • As Task no increases, bound 69% Dr. Hugh Melvin, Dept. of IT, NUI, G 16

Utilisation Bound for RM Alg. 1 0. 83 0. 78 0. 69 0. 73 Bound 1 2 3 4 5 6 10 ∞ No of Tasks Dr. Hugh Melvin, Dept. of IT, NUI, G 17

Earliest Deadline First (EDF) • Dynamic priority and preemptable • Ready task whose absolute deadline is the earliest given highest priority • Task priorities are reevaluated when tasks released/completed • EDF is an optimal uniprocessor sched alg • If all tasks are periodic – Task 1…n ; CPU U = – If U <=1, then task set is EDF schedulable Dr. Hugh Melvin, Dept. of IT, NUI, G 30

EDF Example Task e p u T 1 1 4 0. 25 T 2 2 5 0. 4 T 3 5 20 0. 25 All Tasks released at time 0 Overall U = 0. 9 Sequence 1 st inst Task 1 runs 1 st as earliest deadline of 4 1 st inst Task 2 runs to completion 1 st inst Task 3 runs for 1 unit. . note: Deadline is 20. . at Eu=4, Task 1 rel. preempy Task 3 as deadline is 8 2 nd inst Task 1 runs to completion. . at Eu =5, Task 2 released nd 2 inst Task 2 runs to completion as deadline is 10 1 st inst Task 3 runs for 1 unit. . At Eu = 8, Task 1 released preempts Task 3 3 rd inst Task 1 runs to completion 1 st inst Task 3 runs for 1 unit. . At Eu =10, Task 2 rel. . Preempts task 3 as deadline is 15. . At Eu =12, Task 1 runs as deadline 16 < 20 At Eu =15 Task 2 released and runs with deadline 20 At Eu =16 Task 1 rel with deadline 20 no preemption 1 2 2 3 1 3 2 2 1 3 3 2 2 1 1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Execution Units Eu Dr. Hugh Melvin, Dept. of IT, NUI, G 31

EDF Example Task e p u T 1 1 3 0. 33 T 2 2 5 0. 4 T 3 2 10 0. 2 All Tasks released at time 0 Overall U = 0. 93 Sequence 1 st inst Task 1 runs 1 st as earliest deadline of 3 1 st inst Task 2 runs to completion. . At Eu=3, 2 nd inst Task 1 rel & runs as deadline is 6 1 st inst Task 3 runs. . at Eu=5, Task 2 rel. doesn’t preempt Task 3 as deadline is also 10. . Task 3 completes. . At Eu = 6, Task 1 released deadline 9 runs. . At Eu = 7 Task 2 runs to completion. . At Eu =9, Task 1 rel deadline 12. . runs. . At Eu =10, Task 3 rel. deadline 20, Task 2 also rel with deadline 15 Task 2 runs. . At Eu =15, Task 1 and 2 released with deadlines 16, 20 Task 1 runs. . At Eu=16, Task 2 runs. . At Eu = 18 Task 1 rel with deadline 21 but Task 3 has deadline 20 Task 3 runs 1 2 2 1 3 3 1 2 2 3 3 1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Execution Units Eu Dr. Hugh Melvin, Dept. of IT, NUI, G 32

EDF vs RM • With RM, priorities fixed – Lowest period tasks guaranteed proc time • Only higher period tasks will miss deadlines – In overload conditions, same lower priority tasks lose out – Bound on CPU utilisation must be considered • EDF, dynamic priority – – More flexible. . Less predictable All tasks may miss deadlines Schedulable if CPU U <=1 May react poorly to overload condition Dr. Hugh Melvin, Dept. of IT, NUI, G 33

Conventional OS Scheduling • Time Sharing Emphasis – Balance needs of interactive and CPU intensive traffic • Complex alg – Constantly adjust process priority • CPU intensive processes lose priority as they run • As a process waits, priority is increased will get CPU slot sooner • Unix command nice – nice -20 process_high_priority – nice +20 process_low_priority – nice will work but underlying sched alg unchanged • Windows Real. Time Threads – Somewhat equivalent Dr. Hugh Melvin, Dept. of IT, NUI, G 34

POSIX. 4 RT Scheduling • Defines two main scheduling policies – SCHED_FIFO and SCHED_RR • Each have attributes – Also have SCHED_OTHER – Currently a single attribute = priority struct sched_param{ int sched_priority; } – Eg. Could implement EDF by extending structure to include • struct timespec sched_deadline; • struct timespec sched_timerequired; Dr. Hugh Melvin, Dept. of IT, NUI, G 35

POSIX. 4 RT Scheduling • SCHED_FIFO – Simple priority based preemptive scheduler – Most common in RTS – FIFO used to schedule processes within each priority level – If no other process exists at higher priority, process runs until complete • Next process at that priority (if present) then allocated CPU • Highest priority process guaranteed processor time Dr. Hugh Melvin, Dept. of IT, NUI, G 36

POSIX. 4 RT Scheduling • SCHED_RR – Round robin used to timeslice among processes at same priority level – System provided timeslice – Use for lower priority tasks Dr. Hugh Melvin, Dept. of IT, NUI, G 37

POSIX. 4 RT Scheduling • Setting scheduling policy and attribute #include <sched. h> struct sched_param scheduling_parameters; int scheduling_policy; int i; scheduling_parameters. sched_priority=17; i=sched_setscheduler(getpid( ), SCHED_FIFO, &scheduling_parameters); • getpid( ) used to determine process ID – Process set to FIFO, priority 17 Dr. Hugh Melvin, Dept. of IT, NUI, G 38

POSIX. 4 RT Scheduling • Process priority ranges differ among OS – Need this info before setting priority level int sched_rr_min, sched_rr_max; int sched_fifo_min, sched_fifo_max; sched_rr_min=sched_get_priority_min(SCHED_RR); sched_rr_max=sched_get_priority_max(SCHED_RR); sched_fifo_min=sched_get_priority_min(SCHED_FIFO); sched_fifo_max=sched_get_priority_max(SCHED_FIFO); – Eg. 256 priority levels • FIFO range 128 -255 • RR range 0 -127 Dr. Hugh Melvin, Dept. of IT, NUI, G 39

POSIX. 4 RT Scheduling Eg. #include<sched. h> int i; struct sched_param my_sched_params; // determine max FIFO priority level my_sched_params. sched_priority= sched_get_priority_max(SCHED_FIFO); // Set priority i=sched_setparam(getpid (), &my_sched_params); Dr. Hugh Melvin, Dept. of IT, NUI, G 40