RTS: Kernel Design and Cyclic Executives Chapter 4 Amrita-UB-MSES-2013-6 5/11/2013

Kernel & Device drivers Servers (application ~, web ~, component ~) Shell XWin Thread lib ftp User applications System call interface Process, memory, file system, network managers. Kernel Device drivers Hardware/controller Devices Amrita-UB-MSES-2013-6 5/11/2013

Task characteristics of real workload Each task Ti is characterized by the following temporal parameters: Precedence constraints: specify any tasks need to precede other tasks. Release or arrival time: ri,j: jth instance of ith task Phase Φi: release time of first instant of ith task Response time: time between activation and completion Absolute deadline: instant by which task must complete Relative deadline: maximum allowable response time Period Pi: maximum length of intervals between the release times of consecutive tasks. Execution time: the maximum amount of time required to complete a instance of the task assuming all the resources are available. Amrita-UB-MSES-2013-6 5/11/2013

Simple kernels Polled loop: Say a kernel needs to process packets that are transferred into the DMA and a flag is set after transfer: for(;;) { if (packet_here) { process_data(); packet_here=0; } Excellent for handling high-speed data channels, a processor is dedicated to handling the data channel. Disadvantage: cannot handle bursts Amrita-UB-MSES-2013-6 5/11/2013

Simple kernels: cyclic executives Illusion of simultaneity by taking advantage of relatively short processes in a continuous loop: for(;;) { process_1(); process_2(); process_3(); … process_n(); } Different rate structures can be achieved by repeating tasks in the list: Amrita-UB-MSES-2013-6 5/11/2013

Cyclic Executives: Example: Interactive games Space invaders: for(;;) { check_for_keypressed(); move_aliens(); check_collision(); update_screen(); } check_keypressed() checks for three button pressings: move tank left or right and fire missiles. If the schedule is carefully constructed we could achieve a very efficient game program with a simple kernel as shown above. Amrita-UB-MSES-2013-6 5/11/2013

Finite state automata and Co-routine based kernels void process_a(void){ for(;;) { switch (state_a) { case 1: phase_a1(); | case 2: phase_a2(); | …. case n: phase_an();}}} void process_b(void){ switch (state_b) { case 1: phase_b1(); | case 2: phase_b2(); | case n: phase_bn();}}} state_a and state_b are state counters; Communication between coroutines thru’ global variables; Example: the famous CICS from IBM : Customer Information Control System IBM’s OS/2 uses this in Windows presentation management. Amrita-UB-MSES-2013-6 5/11/2013

Interrupt driven systems Main program is a simple loop. Various tasks in the system are schedules via software or hardware interrupts; Dispatching performed by interrupt handling routines. Hardware and software interrupts. Hardware: asynchronous Software: typically synchronous Executing process is suspended, state and context saved and control is transferred to ISR (interrupt service routine) Amrita-UB-MSES-2013-6 5/11/2013

Interrupt driven systems: code example void main() { init(); while(TRUE); } void int1(void){ save (context); task1(); retore (context);} restore (context);} Foreground/background systems is a variation of this where main does some useful task in the background; Amrita-UB-MSES-2013-6 5/11/2013

Process scheduling Scheduling is a very important function in a real-time operating system. Two types: pre-run-time and run-time Pre-run-time scheduling: create a feasible schedule offline to meet time constraints, guarantee execution order of processes, and prevents simultaneous accesses to shared resources. Run-time scheduling: allows events to interrupt processes, on demand allocation of resources , and used complex run-time mechanisms to meet time constraints. Amrita-UB-MSES-2013-6 5/11/2013

More on Cyclic Executives Simple loop cyclic executive Frame/slots Table-based predetermined schedule cyclic executive Periodic, aperiodic and interrupt-based task Lets design a cyclic-executive with multiple periodic tasks. 11 Amrita-UB-MSES-2013-6 5/11/2013

The basic systems Several functions are called in a prearranged sequence Some kind of cooperative scheduling You a have a set of tasks and a scheduler that schedules these tasks Types of tasks: base tasks (background), interrupt tasks, clock tasks Frame of slots, slots of cycles, each task taking a cycle, burn tasks to fill up the left over cycles in a frame. 12 Amrita-UB-MSES-2013-6 5/11/2013

Cyclic Executive Design 1 (pages 81-87) Base tasks, clock tasks, interrupt tasks Base: no strict requirements, background activity Clock: periodic with fixed runtime Interrupt: event-driven preemption, rapid response but little processing Design the slots Table-driven cyclic executive Amrita-UB-MSES-2013-6 5/11/2013

Cyclic executive Each task implemented as a function All tasks see global data and share them Cyclic executive for three priority level The execution sequence of tasks within a cyclic executive will NOT vary in any unpredictable manner (such as in a regular fully featured Operating Systems) Clock tasks, clock sched, base tasks, base sched, interrupt tasks Each clock slot executes, clock tasks, at the end a burn task that is usually the base task Study the figures in pages 83-86 of your text Amrita-UB-MSES-2013-6 5/11/2013

RT Cyclic Executive Program Lets examine the code: Identify the tasks Identify the cyclic schedule specified in the form of a table Observe how the functions are specified as table entry Understand the scheduler is built-in Learn how the function in the table are dispatched Amrita-UB-MSES-2013-6 5/11/2013

Implementation of a cyclic executive #include <stdio.h> #include <ctype.h> #include <unistd.h> #include <sys/times.h> #define SLOTX 4 #define CYCLEX 5 #define SLOT_T 5000 int tps,cycle=0,slot=0; clock_t now, then; struct tms n; void one() { printf("Task 1 running\n"); sleep(1); } void two() { printf("Task 2 running\n"); sleep(1); } Amrita-UB-MSES-2013-6 5/11/2013

Implementation (contd.) void three() { printf("Task 3 running\n"); sleep(1); } void four() { printf("Task 4 running\n"); void five() { printf("Task 5 running\n"); Amrita-UB-MSES-2013-6 5/11/2013

Implementation (contd.) void burn() { clock_t bstart = times(&n); while ((( now = times(&n)) - then) < SLOT_T * tps / 1000) { } printf (" brn time = %2.2dms\n\n", (times(&n)-bstart)*1000/tps); then = now; cycle = CYCLEX; } Amrita-UB-MSES-2013-6 5/11/2013

Implementation (contd.) void (*ttable[SLOTX][CYCLEX])() = { {one, two, burn, burn, burn}, {one, three, four, burn, burn}, {one, five, four, burn, burn}}; main() { tps = sysconf(_SC_CLK_TCK); printf("clock ticks/sec = %d\n\n", tps); then = times(&n); while (1) { for (slot=0; slot <SLOTX; slot++) for (cycle=0; cycle<CYCLEX; cycle++) (*ttable[slot][cycle])(); }} Amrita-UB-MSES-2013-6 5/11/2013

Summary The cyclic executive discussed the scheduler is built-in. You can also use clock ticks RTC etc to schedule the tasks In order use the cyclic executive discussed here in other applications simple change table configuration, and rewrite the dummy functions we used. Amrita-UB-MSES-2013-6 5/11/2013