Round robin scheduling example in c

Os Scheduling

INSTRUCTIONS

============

1. OVERVIEW

===========

In this MP, you will write a user-mode thread scheduler. The basic purpose

of a scheduler is to multiplex use of the computer across several threads

of execution. This MP deals with two different scheduling policies: FIFO

and Round Robin. You will implement both, for use in a simple cooperative

multi-threading system. Along the way, you'll also learn about implementing

object-oriented constructs in low-level procedural languages like C.

This assignment consists of implementing the core functionality of the

scheduler (Step 4) and answering 10 questions (Step 5). Code for

Step 4 goes in sched_impl.c and sched_impl.h.

2. THEORY OF OPERATION

======================

The given code in the MP defines the skeleton of a scheduler together with a

parameterized dummy workload. The idea is when you run the MP, you specify

a scheduling policy, scheduler queue size, some number of worker threads to

create, and, optionally, the number of iterations for which the worker

threads should run. The basic code that parses command line arguments and

creates these worker threads is provided in the MP, but you must implement

the core synchronization and scheduling operations.

As provided, the MP only includes the "dummy" scheduling algorithm, which

doesn't even try to do anything. You can run it like this:

make

./scheduler -dummy 0 N # where N is some number of worker threads

All threads run right away regardless of the queue size (even zero!), and

are scheduled by the operating system. The goal of this MP is to create

scheduler implementations which are a bit more controlled and predictable.

For example, once you have completed the MP, the following should work:

./scheduler -fifo 1 2 3

Main: running 2 workers on 1 queue_size for 3 iterations

Main: detaching worker thread 3075984304

Main: detaching worker thread 3065494448

Main: waiting for scheduler 3086474160

Thread 3075984304: in scheduler queue

Thread 3075984304: loop 0

Thread 3075984304: loop 1

Thread 3075984304: loop 2

Thread 3075984304: exiting

Thread 3065494448: in scheduler queue

Thread 3065494448: loop 0

Thread 3065494448: loop 1

Thread 3065494448: loop 2

Thread 3065494448: exiting

Scheduler: done!

The command line options used above specify:

-fifo Use FIFO scheduling policy

1 One thread can be in the scheduler queue at a time

2 Create 2 worker threads

3 Each thread runs for 3 time slices

Here's another example:

./scheduler -rr 10 2 3

Main: running 2 workers on 10 queue_size for 3 iterations

Main: detaching worker thread 3075828656

Main: detaching worker thread 3065338800

Main: waiting for scheduler 3086318512

Thread 3075828656: in scheduler queue

Thread 3065338800: in scheduler queue

Thread 3075828656: loop 0

Thread 3065338800: loop 0

Thread 3075828656: loop 1

Thread 3065338800: loop 1

Thread 3075828656: loop 2

Thread 3065338800: loop 2

Thread 3075828656: exiting

Thread 3065338800: exiting

Scheduler: done!

The command line options used above specify:

-rr Use Round Robin scheduling policy

10 Ten threads can be in the scheduler queue at a time

2 Create 2 worker threads

3 Each thread runs for 3 time slices

Things to observe:

In both examples, the worker threads are created at the beginning of

execution. But in the case with queue size 1, one of the threads has to

wait until the other thread exits before it can enter the scheduler queue

(the "in scheduler queue" messages). Whereas in the case with queue size

10, both threads enter the scheduler queue immediately.

The FIFO policy would actually have basically the same behavior even with a

larger queue size; the waiting worker threads would simply be admitted to

the queue earlier.

The Round Robin scheduling policy alternates between executing the two

available threads, until they run out of work to do.

3. FILE LAYOUT

==============

The MP distribution consists of the following source files:

scheduler.c

Includes the skeleton of a scheduler (sched_proc()) and a

parameterized dummy workload (worker_proc()). The main() function

accepts several parameters specifying the test workload (see

description below). The scheduler relies on a scheduler

implementation (sched_impl_t) to implement the specifics of its

scheduling policy (to be provided by you in sched_impl.[hc])

scheduler.h

Describes the interface to which your scheduler implementation must

adhere. The structures containing function pointers are similar to

Java interfaces or C++ pure virtual base classes. This file

declares that you must define two sched_impl_t structures,

sched_fifo and sched_rr in another file (sched_impl.c).

dummy_impl.c

Implements the dummy scheduling algorithm, which just lets the OS

schedule all threads, regardless of queue size.

sched_impl.h (define your data structures here)

This is where you will define the data structures stored per

scheduler instance (struct sched_queue) and per worker thread

(struct thread_info). This will likely include synchronization

constructs like semaphores and mutexes, and a list of threads

available to be scheduled.

sched_impl.c (implement your code here)

This is where you will define the functions implementing the core

behavior of the scheduler, including the FIFO and Round Robin

scheduling policies. The only way functions defined in this file

are made available to the main program (scheduler.c) is by placing

function pointers in the sched_impl_t structures sched_fifo and

sched_rr.

list.h

Defines the basic operations on a bidirectional linked list data

structure. The elements of the list, of type list_elem_t, include

a void *datum where you can store pointers to whatever kind of

data you like. You don't have to use this linked list library,

but it will probably come in handy.

list.c

Implements the linked list operations.

smp4_tests.c

testrunner.c

testrunner.h

Test harness, defines test cases for checking your MP solution.

Please take a look at the source files and familiarize yourself with how

they work. Think about how structures containing function pointers compare

to classes and virtual methods in C++. If you'd like to learn more, read

about the virtual function table in C++. The struct containing function

pointers technique employed in this MP is also used by C GUI libraries like

GTK+ and to define the operations of loadable modules, such as file systems,

within the Linux kernel.

4. PROGRAMMING

==============

Now you're ready to implement the core of the scheduler, including the FIFO

and Round Robin scheduling algorithms. For this purpose, you should only

modify sched_impl.h and sched_impl.c. Please see scheduler.h for the

descriptions of what functions you must implement. You are free to put

whatever you want in the thread_info and sched_queue structures. Note that

the only way that the functions you implement are made available to the main

program is through the sched_impl_t structures sched_fifo and sched_rr. See

dummy_impl.c for a completed example of how to fill in a sched_impl_t.

Suggested approach:

4.1 Create stub versions of all of the functions you will need to implement

in sched_impl.c, and statically initialize sched_fifo and sched_rr.

4.2 Figure out how you will implement the scheduler queue, add the

appropriate fields to struct sched_queue, and fill in the appropriate

queue-related operations in the functions you created in (4.1).

Remember that we provide a doubly-linked list in list.[hc].

4.3 Implement scheduler queue admission control, so that only the requested

number of threads can be in the scheduler queue at once. Create the

appropriate synchronization constructs to prevent threads not in the

queue from executing (look at the implementation of worker threads in

scheduler.c:worker_proc()).

4.4 Implement the queue operations for selecting the next thread to execute.

This will be different for FIFO vs. Round Robin scheduling.

4.5 Add in synchronization constructs to make sure only the selected thread

executes at any given time.

4.6 Fill in any gaps that might remain.

When you think you're done, you can test your program using the command

"make test". For more thorough testing, the fifo_var and rr_var tests

accept queue_size, num_workers, and num_iterations arguments just like the

main program (but is mandatory for the test case):

./scheduler -test fifo_var

./scheduler -test rr_var

5. QUESTIONS

============

Q 1 What are some pros and cons of using the struct of function pointers

approach as we did in the MP to link different modules? Does it

significantly affect performance? Give some examples of when you would

and wouldn't use this approach, and why.

Q 2 Briefly describe the synchronization constructs you needed to implement

this MP--i.e., how you mediated admission of threads to the scheduler

queue and how you made sure only the scheduled thread would run at any

given time.

Q 3 Why is the dummy scheduling implementation provided potentially

unsafe (i.e. could result in invalid memory references)? How does

your implementation avoid this problem?

Q 4 When using the FIFO or Round Robin scheduling algorithm, can

sched_proc() ever "miss" any threads and exit early before all threads

have been scheduled and run to completion? If yes, give an example; if

no, explain why not.

Q 5 Why are the three variables in scheduler.h declared 'extern'? What

would happen if they were not declared 'extern'? What would happen

if they were not declared without the 'extern' in any file?

Q 6 Describe the behavior of exit_error() function in scheduler.c. Why

does exit_error() not use errno?

Q 7 Does it matter whether the call to sched_ops->wait_for_queue(queue) in

sched_proc() actually does anything? How would it affect correctness

if it just returned right away? How about performance?

Q 8 Explain how worker_proc() is able to call the appropriate

implementation of wait_for_cpu() corresponding to the scheduling policy

selected by the user on the command line. Start from main() and

briefly explain each step along the way.

Q 9 Is it possible that a worker thread would never proceed past the call to

wa->ops->wait_for_cpu(&wa->info) when using one of the scheduling

policies implemented in this MP?

Q 10 Explain how you would alter the program to demonstrate the "convoy"

effect, when a large compute bound job that never yields to another

thread slows down all other jobs in a FIFO scheduled system? See Page

402, Stallings, the paragraph starting "Another difficulty with FCFS is

that it tends to favor processor-bound processes over I/O bound

processes". Why is it difficult to show the benefits of Round Robin

scheduling in this case using the current implementation in the machine

problem?

...................................................................................................................................

SMP5: Scheduler with Signals (PART 2)

============================

This MP is a variation of SMP4.

In the last MP, we built a simulated OS process scheduler. The scheduler can

hold only a certain number of processes (workers) at one time. Once the process

has been accepted into the scheduler, the scheduler decides in what order the

processes execute. We implemented two scheduling algorithms: FIFO and Round

Robin.

In this MP, we are to simulate a time-sharing system by using signals and

timers. We will only implement the Round Robin algorithm. Instead of using

iterations to model the concept of "time slices" (as in the last MP), we use

interval timers. The scheduler is installed with an interval timer. The timer

starts ticking when the scheduler picks a thread to use the CPU which in turn

signals the thread when its time slice is finished thus allowing the scheduler

to pick another thread and so on. When a thread has completely finished its work

it leaves the scheduler to allow a waiting thread to enter. Please note that in

this MP, only the timer and scheduler send signals. The threads passively handle

the signals without signaling back to the scheduler.

The program takes a number of arguments. Arg1 determines the number of jobs

(threads in our implementation) created; arg2 specifies the queue size of the

scheduler. Arg3 through argN gives the duration (the required time slices to

complete a job) of each job. Hence if we create 2 jobs, we should supply arg3

and arg4 for the required duration. You can assume that the autograder will

always supply the correct number of arguments and hence you do not have to

detect invalid input.

Here is an example of program output, once the program is complete:

% scheduler 3 2 3 2 3

Main: running 3 workers with queue size 2 for quanta:

3 2 3

Main: detaching worker thread 3075926960.

Main: detaching worker thread 3065437104.

Main: detaching worker thread 3054947248.

Main: waiting for scheduler 3086416816.

Scheduler: waiting for workers.

Thread 3075926960: in scheduler queue.

Thread 3075926960: suspending.

Thread 3065437104: in scheduler queue.

Thread 3065437104: suspending.

Scheduler: scheduling.

Scheduler: resuming 3075926960.

Thread 3075926960: resuming.

Scheduler: suspending 3075926960.

Scheduler: scheduling.

Scheduler: resuming 3065437104.

Thread 3065437104: resuming.

Thread 3075926960: suspending.

Scheduler: suspending 3065437104.

Scheduler: scheduling.

Scheduler: resuming 3075926960.

Thread 3075926960: resuming.

Thread 3065437104: suspending.

Scheduler: suspending 3075926960.

Scheduler: scheduling.

Scheduler: resuming 3065437104.

Thread 3065437104: resuming.

Thread 3075926960: suspending.

Scheduler: suspending 3065437104.

Thread 3065437104: leaving scheduler queue.

Scheduler: scheduling.

Scheduler: resuming 3075926960.

Thread 3075926960: resuming.

Thread 3065437104: terminating.

Thread 3054947248: in scheduler queue.

Thread 3054947248: suspending.

Scheduler: suspending 3075926960.

Thread 3075926960: leaving scheduler queue.

Scheduler: scheduling.

Scheduler: resuming 3054947248.

Thread 3054947248: resuming.

Thread 3075926960: terminating.

Scheduler: suspending 3054947248.

Scheduler: scheduling.

Scheduler: resuming 3054947248.

Thread 3054947248: suspending.

Thread 3054947248: resuming.

Scheduler: suspending 3054947248.

Scheduler: scheduling.

Scheduler: resuming 3054947248.

Thread 3054947248: suspending.

Thread 3054947248: resuming.

Scheduler: suspending 3054947248.

Thread 3054947248: leaving scheduler queue.

Thread 3054947248: terminating.

The total wait time is 12.062254 seconds.

The total run time is 7.958618 seconds.

The average wait time is 4.020751 seconds.

The average run time is 2.652873 seconds.

The goal of this MP is to help you understand (1) how signals and timers work,

and (2) how to evaluate the performance of your program. You will first

implement the time-sharing system using timers and signals. Then, you will

evaluate the overall performance of your program by keeping track of how long

each thread is idle, running, etc.

The program will use these four signals:

SIGALRM: sent by the timer to the scheduler, to indicate another time

quantum has passed.

SIGUSR1: sent by the scheduler to a worker, to tell it to suspend.

SIGUSR2: sent by the scheduler to a suspended worker, to tell it to resume.

SIGTERM: sent by the scheduler to a worker, to tell it to cancel.

You will need to set up the appropriate handlers and masks for these signals.

You will use these functions:

clock_gettime

pthread_sigmask

pthread_kill

sigaction

sigaddset

sigemptyset

sigwait

timer_settime

timer_create

Also, make sure you understand how the POSIX:TMR interval timer works.

There are two ways you can test your code. You can run the built-in grading

tests by running "scheduler -test -f0 rr". This runs 5 tests, each of which can

be run individually. You can also test you program with specific parameters by

running "scheduler -test gen ..." where the ellipsis contains the parameters you

would pass to scheduler.

Programming

===========

Part I: Modify the scheduler code (scheduler.c)

-----------------------------------------------

We use the scheduler thread to setup the timer and handle the scheduling for the

system. The scheduler handles the SIGALRM events that come from the timer, and

sends out signals to the worker threads.

Step 1.

Modify the code in init_sched_queue() function in scheduler.c to initialize the

scheduler with a POSIX:TMR interval timer. Use CLOCK_REALTIME in timer_create().

The timer will be stored in the global variable "timer", which will be started

in scheduler_run() (see Step 4 below).

Step 2.

Implement setup_sig_handlers(). Use sigaction() to install signal handlers for

SIGALRM, SIGUSR1, and SIGTERM. SIGALRM should trigger timer_handler(), SIGUSR1

should trigger suspend_thread(), and SIGTERM should trigger cancel_thread().

Notice no handler is installed for SIGUSR2; this signal will be handled

differently, in step 8.

Step 3.

In the scheduler_run() function, start the timer. Use timer_settime(). The

time quantum (1 second) is given in scheduler.h. The timer should go off

repeatedly at regular intervals defined by the timer quantum.

In Round-Robin, whenever the timer goes off, the scheduler suspends the

currently running thread, and tells the next thread to resume its operations

using signals. These steps are listed in timer_handler(), which is called every

time the timer goes off. In this implementation, the timer handler makes use of

suspend_worker() and resume_worker() to accomplush these steps.

Step 4.

Complete the suspend_worker() function. First, update the info->quanta value.

This is the number of quanta that remain for this thread to execute. It is

initialized to the value passed on the command line, and decreases as the thread

executes. If there is any more work for this worker to do, send it a signal to

suspend, and update the scheduler queue. Otherwise, cancel the thread.

Step 5.

Complete the cancel_worker() function by sending the appropriate signal to the

thread, telling it to kill itself.

Step 6.

Complete the resume_worker() function by sending the appropriate signal to the

thread, telling it to resume execution.

Part II: Modify the worker code (worker.c)

------------------------------------------

In this section, you will modify the worker code to correctly handle the signals

from the scheduler that you implemented in the previous section.

You need to modify the thread functions so that it immediately suspends the

thread, waiting for a resume signal from the scheduler. You will need to use

sigwait() to force the thread to suspend itself and wait for a resume signal.

You need also to implement a signal handler in worker.c to catch and handle the

suspend signals.

Step 7.

Modify start_worker() to (1) block SIGUSR2 and SIGALRM, and (2) unblock SIGUSR1

and SIGTERM.

Step 8.

Implement suspend_thread(), the handler for the SIGUSR1 signal. The

thread should block until it receives a resume (SIGUSR2) signal.

Part III: Modify the evaluation code (scheduler.c)

--------------------------------------------------

This program keeps track of run time, and wait time. Each thread saves these

two values regarding its own execution in its thread_info_t. Tracking these

values requires also knowing the last time the thread suspended or resumed.

Therefore, these two values are also kept in thread_info_t. See scheduler.h.

In this section, you will implement the functions that calculate run time and

wait time. All code that does this will be in scheduler.c. When the program

is done, it will collect all these values, and print out the total and average

wait time and run time. For your convenience, you are given a function

time_difference() to compute the difference between two times in microseconds.

Step 9.

Modify create_workers() to initialize the various time variables.

Step 10.

Implement update_run_time(). This is called by suspend_worker().

Step 11.

Implement update_wait_time(). This is called by resume_worker().

Questions

==========

Question 1.

Why do we block SIGUSR2 and SIGALRM in worker.c? Why do we unblock SIGUSR1 and

SIGTERM in worker.c?

Question 2.

We use sigwait() and sigaction() in our code. Explain the difference between the

two. (Please explain from the aspect of thread behavior rather than syntax).

Question 3.

When we use POSIX:TMR interval timer, we are using relative time. What is the

alternative? Explain the difference between the two.

Question 4.

Look at start_worker() in worker.c, a worker thread is executing within an

infinite loop at the end. When does a worker thread terminate?

Question 5.

When does the scheduler finish? Why does it not exit when the scheduler queue

is empty?

Question 6.

After a thread is scheduled to run, is it still in the sched_queue? When is it

removed from the head of the queue? When is it removed from the queue completely?

Question 7.

We've removed all other condition variables in SMP4, and replaced them with a

timer and signals. Why do we still use the semaphore queue_sem?

Question 8.

What's the purpose of the global variable "completed" in scheduler.c? Why do we

compare "completed" with thread_count before we wait_for_queue() in

next_worker()?

Question 9.

We only implemented Round Robin in this SMP. If we want to implement a FIFO

scheduling algorithm and keep the modification as minimum, which function in

scheduler.c is the one that you should modify? Briefly describe how you would

modify this function.

Question 10.

In this implementation, the scheduler only changes threads when the time quantum

expires. Briefly explain how you would use an additional signal to allow the

scheduler to change threads in the middle of a time quantum. In what situations

would this be useful?

assign3/assign3_part2/list.c
#include #include "list.h" /* list helper functions */ int list_size(thread_info_list *list) { int cnt = 0; if (!list) return -1; pthread_mutex_lock(&list->lock); list_elem *le = list->head; while (le) { cnt++; le = le->next; } pthread_mutex_unlock(&list->lock); return cnt; } int list_insert_head(thread_info_list *list, list_elem *new) { if (!list || !new) return -1; pthread_mutex_lock(&list->lock); new->next = list->head; new->prev = 0; if (new->next) { new->next->prev = new; } list->head = new; if (list->tail == 0) { list->tail = new; } pthread_mutex_unlock(&list->lock); return 0; } int list_insert_tail(thread_info_list *list, list_elem *new) { if (!list || !new) return -1; pthread_mutex_lock(&list->lock); new->prev = list->tail; new->next = 0; if (new->prev) { new->prev->next = new; } list->tail = new; if (list->head == 0) { list->head = new; } pthread_mutex_unlock(&list->lock); return 0; } int list_remove(thread_info_list *list, list_elem *old) { if (!old || !list) return -1; pthread_mutex_lock(&list->lock); if (old->next) { old->next->prev = old->prev; } if (old->prev) { old->prev->next = old->next; } if (list->tail == old) { list->tail = old->prev; } if (list->head == old) { list->head = old->next; } old->next = old->prev = 0; pthread_mutex_unlock(&list->lock); return 0; } void print_list(thread_info_list *list) { pthread_mutex_lock(&list->lock); list_elem *le = list->head; while (le) { printf("0x%X,", (unsigned int)le->info); le = le->next; } pthread_mutex_unlock(&list->lock); printf("\n"); }

__MACOSX/assign3/assign3_part2/._list.c
assign3/assign3_part2/worker.c
#include #include #include #include #include #include "scheduler.h" /******************************************************************************* * * Implement these functions. * ******************************************************************************/ /* Handler for SIGTERM signal */ void cancel_thread() { printf("Thread %u: terminating.\n", (unsigned int)pthread_self()); /* signal that done in queue */ sem_post(&queue_sem); pthread_exit(NULL); } /* TODO: Handle the SIGUSR1 signal */ void suspend_thread() { printf("Thread %u: suspending.\n", (unsigned int)pthread_self()); /*add your code here to wait for a resume signal from the scheduler*/ printf("Thread %u: resuming.\n",(unsigned int) pthread_self()); } /******************************************************************************* * * * ******************************************************************************/ /* * waits to gain access to the scheduler queue. */ static int enter_scheduler_queue(thread_info_t *info) { /* * wait for available room in queue. * create a new list entry for this thread * store this thread info in the new entry. */ sem_wait(&queue_sem); list_elem *item = (list_elem*)malloc(sizeof(list_elem)); info->le = item; item->info = info; item->prev = 0; item->next = 0; list_insert_tail(&sched_queue, item); return 0; } /* * leaves the scheduler queue */ void leave_scheduler_queue(thread_info_t *info) { printf("Thread %lu: leaving scheduler queue.\n", info->thrid); /* * remove the given worker from queue * clean up the memory that we malloc'd for the list * clean up the memory that was passed to us */ list_remove(&sched_queue, info->le); free(info->le); free(info); } /* * Initialize thread, enter scheduling queue, and execute instructions. * arg is a pointer to thread_info_t */ void *start_worker(void *arg) { thread_info_t *info = (thread_info_t *) arg; float calc = 0.8; int j = 0; /* TODO: Block SIGALRM and SIGUSR2. */ /* TODO: Unblock SIGUSR1 and SIGTERM. */ /* compete with other threads to enter queue. */ if (enter_scheduler_queue(info)) { printf("Thread %lu: failure entering scheduler queue - %s\n", info->thrid, strerror(errno)); free (info); pthread_exit(0); } printf("Thread %lu: in scheduler queue.\n", info->thrid); suspend_thread(); while (1) { /* do some meaningless work... */ for (j = 0; j < 10000000; j++) { calc = 4.0 * calc * (1.0 - calc); } } }

__MACOSX/assign3/assign3_part2/._worker.c
assign3/assign3_part2/Makefile
CC = gcc CCOPTS = -c -g -Wall LINKOPTS = -g -pthread -Wall EXEC=scheduler OBJECTS=scheduler.o worker.o list.o smp5_tests.o testrunner.o all: $(EXEC) $(EXEC): $(OBJECTS) $(CC) $(LINKOPTS) -o $@ $^ -lrt %.o:%.c $(CC) $(CCOPTS) -o $@ $^ clean: - $(RM) $(EXEC) - $(RM) $(OBJECTS) - $(RM) *~ - $(RM) core.* test: scheduler - ./scheduler -test -f0 rr - killall -q -KILL scheduler; true pretty: indent *.c *.h -kr

__MACOSX/assign3/assign3_part2/._Makefile
assign3/assign3_part2/scheduler.h
#ifndef __SCHEDULER_H_ #define __SCHEDULER_H_ #include #include #include "list.h" #define QUANTUM 1 /* typedefs */ typedef struct thread_info { pthread_t thrid; int quanta; list_elem* le; /*added for evalution bookkeeping*/ struct timespec suspend_time; struct timespec resume_time; long wait_time; long run_time; } thread_info_t; /* functions */ void *start_worker(void *); long time_difference(const struct timespec*, const struct timespec*); int smp5_main(int argc, const char** argv); /* shared variables */ extern sem_t queue_sem; /* semaphore for scheduler queue */ extern thread_info_list sched_queue; /* list of current workers */ #endif /* __SCHEDULER_H_ */

__MACOSX/assign3/assign3_part2/._scheduler.h
assign3/assign3_part2/smp5_tests.c
/*************** YOU SHOULD NOT MODIFY ANYTHING IN THIS FILE ***************/ #define _GNU_SOURCE #include #undef _GNU_SOURCE #include #include #include #include #include #include #include "testrunner.h" #include "list.h" #include "scheduler.h" #include "worker.h" //#define quit_if(cond) do {if (cond) exit(EXIT_FAILURE);} while(0) #define quit_if(cond) do {if (cond) {printf("Line %d.",__LINE__);exit(EXIT_FAILURE);}} while(0) void args_to_nums(int argc, const char **argv, int *num_workers, int *queue_size, int **quanta) { int i; quit_if(argc < 4); *num_workers = atoi(argv[1]); *queue_size = atoi(argv[2]); *quanta = malloc(*num_workers * sizeof(int)); quit_if(*quanta == NULL); for(i=3;i 0) { //waitpid(fork_pid, 0, 0); //} else { //fprintf(stderr, "run_test: fork() error\n"); //} } int test_output(FILE *stream, int nw, int qs, int *q) { int queue_size, queue_index; int num_workers, worker_index; int rv, in_queue, term, susp; unsigned long *queue, *workers, tid, prev, newwork, dummyl; int *remaining, *quanta; char dummyc; float tot_wait, tot_run, ave_wait, ave_run; int my_run, my_wait; rv = fscanf(stream,"Main: running %d workers with queue size %d for quanta:\n",&num_workers, &queue_size); quit_if(rv != 2 || num_workers != nw || queue_size != qs); queue = malloc(queue_size*sizeof(long)); workers = malloc(num_workers*sizeof(long)); quanta = malloc(num_workers*sizeof(int)); remaining = malloc(queue_size*sizeof(int)); for(worker_index=0;worker_index 0) { while(!queue[queue_index]) queue_index= (queue_index+1)%queue_size; quit_if(fscanf(stream, "Scheduler: scheduling.%c",&dummyc)!=1 || dummyc != '\n'); quit_if(fscanf(stream, "Scheduler: resuming %lu.\n",&tid) != 1); quit_if( tid != queue[queue_index]); if (prev == tid) { if(term) { quit_if(fscanf(stream, "Thread %lu: terminating.\n",&tid) != 1 || tid != prev); } else if (susp){ quit_if(fscanf(stream, "Thread %lu: suspending.\n",&tid) != 1); quit_if( tid != prev); } quit_if(fscanf(stream, "Thread %lu: resuming.\n",&tid) != 1); quit_if(tid != queue[queue_index]); } else { quit_if(fscanf(stream, "Thread %lu: resuming.\n",&tid) != 1 || tid != queue[queue_index]); if(term) { if(queue_size == 1) quit_if(fscanf(stream, "Scheduler: waiting for workers.%c",&dummyc)!=1 || dummyc!='\n'); quit_if(fscanf(stream, "Thread %lu: terminating.\n",&tid) != 1 || tid != prev); if (in_queue == queue_size) { quit_if(fscanf(stream, "Thread %lu: in scheduler queue.\n",&tid)!=1||tid != newwork); quit_if(fscanf(stream, "Thread %lu: suspending.\n",&tid)!=1 || tid!=newwork); } } else if (susp && in_queue>1){ quit_if(fscanf(stream, "Thread %lu: suspending.\n",&tid) != 1); quit_if( tid != prev); prev = tid; } } quit_if(fscanf(stream, "Scheduler: suspending %lu.\n",&tid) != 1); quit_if(tid != queue[queue_index]); if(!--remaining[queue_index]) { quit_if(fscanf(stream, "Thread %lu: leaving scheduler queue.\n",&tid)!=1 || tid != queue[queue_index]); term = 1; if(worker_index < num_workers) { queue[queue_index] = workers[worker_index]; remaining[queue_index] = quanta[worker_index]; newwork = workers[worker_index]; worker_index++; if(queue_size == 1) { prev = tid; quit_if(fscanf(stream, "Scheduler: waiting for workers.%c",&dummyc)!=1 || dummyc!='\n'); quit_if(fscanf(stream, "Thread %lu: terminating.\n",&tid) != 1 || tid != prev); quit_if(fscanf(stream, "Thread %lu: in scheduler queue.\n",&tid)!= 1 || tid != newwork); quit_if(fscanf(stream, "Thread %lu: suspending.\n",&tid)!= 1 || tid != newwork); term = 0; susp = 0; my_wait++; } } else { queue[queue_index] = 0; in_queue--; } } else { term = 0; susp = 1; } prev = tid; my_run++; my_wait += in_queue+(num_workers-worker_index)-1+term; queue_index= (queue_index+1)%queue_size; } quit_if(fscanf(stream, "Th%c",&dummyc) != 1); if (dummyc=='r') { quit_if(fscanf(stream, "ead %lu: terminating.\nThe",&tid)!=1 || tid != prev); } quit_if(fscanf(stream, " total wait time is %f seconds.\n",&tot_wait) != 1); quit_if(fscanf(stream, "The total run time is %f seconds.\n",&tot_run) != 1); quit_if(fscanf(stream, "The average wait time is %f seconds.\n",&ave_wait) != 1); quit_if(fscanf(stream, "The average run time is %f seconds.\n",&ave_run) != 1); if (dummyc=='e') quit_if(fscanf(stream, "Thread %lu: terminating.\nThe",&tid) != 1|| tid != prev); quit_if(abs(tot_wait-my_wait)>1); quit_if(abs(tot_run-my_run)>1); quit_if(abs(tot_wait/num_workers-ave_wait)>.5); quit_if(abs(tot_run/num_workers-ave_run)>.5); return 0; } int general_test(int argc, const char **argv) { FILE *f; int nw, qs, *q; run_test(argc,argv); f = fopen("smp5.out","r"); args_to_nums(argc,argv,&nw,&qs,&q); test_output(f,nw,qs,q); return EXIT_SUCCESS; } int specific_test(int nw, int qs, int *q) { FILE *f; int argc; char **argv; nums_to_args(nw,qs,q,&argc,&argv); run_test(argc,(const char **)argv); f = fopen("smp5.out","r"); test_output(f,nw,qs,q); return EXIT_SUCCESS; } int test_3_1_2_2_2() { int q[3] = {2,2,2}; return specific_test(3,1,q); } int test_2_2_2_2() { int q[2]={2,2}; return specific_test(2,2,q); } int test_5_7_1_2_1_2_1() { int q[5] = {1,2,1,2,1}; return specific_test(5,7,q); } int test_4_1_1_2_3_4() { int q[4] = {1,2,3,4}; return specific_test(4,1,q); } int test_3_3_4_3_2() { int q[3] = {4,3,2}; return specific_test(3,3,q); } /* * Main entry point for SMP% test harness */ int run_smp5_tests(int argc, const char **argv) { /* Tests can be invoked by matching their name or their suite name * or 'all' */ testentry_t tests[] = { {"test_3_1_2_2_2", "rr", test_3_1_2_2_2}, {"test_2_2_2_2", "rr", test_2_2_2_2}, {"test_5_7_1_2_1_2_1", "rr", test_5_7_1_2_1_2_1}, {"test_4_1_1_2_3_4", "rr", test_4_1_1_2_3_4}, {"test_3_3_4_3_2", "rr", test_3_3_4_3_2}, {"general", "gen", general_test} }; int result = run_testrunner(argc, argv, tests, sizeof(tests) / sizeof(testentry_t)); unlink("smp5.out"); return result; } /* The real main function. */ int main(int argc, const char **argv) { if (argc > 1 && !strcmp(argv[1], "-test")) { return run_smp5_tests(argc - 1, argv + 1); } else { return smp5_main(argc, argv); } }

__MACOSX/assign3/assign3_part2/._smp5_tests.c
assign3/assign3_part2/testrunner.h
/*************** YOU SHOULD NOT MODIFY ANYTHING IN THIS FILE ***************/ typedef int (*test_fp) (int, const char **); typedef struct { const char *name; const char *suite; test_fp test_function; } testentry_t; int run_testrunner(int argc, const char **argv, testentry_t * entries, int entry_count); void set_testrunner_default_timeout(int s); void set_testrunner_timeout(int s);

__MACOSX/assign3/assign3_part2/._testrunner.h
assign3/assign3_part2/worker.h
#ifndef __WORKER_H_ #define __WORKER_H_ void cancel_thread(); void suspend_thread(); void leave_scheduler_queue(thread_info_t*); #endif

__MACOSX/assign3/assign3_part2/._worker.h
assign3/assign3_part2/list.h
#ifndef __LIST_H_ #define __LIST_H_ #include typedef struct list_elem { struct list_elem *prev; struct list_elem *next; void *info; } list_elem; typedef struct thread_info_list { list_elem *head; list_elem *tail; pthread_mutex_t lock; } thread_info_list; int list_size(thread_info_list *list); int list_insert_head(thread_info_list *list, list_elem *new); int list_insert_tail(thread_info_list *list, list_elem *new); int list_remove(thread_info_list *list, list_elem *new); void print_list(thread_info_list *list); #endif /* __LIST_H_ */

__MACOSX/assign3/assign3_part2/._list.h
assign3/assign3_part2/README.txt
SMP5: Scheduler with Signals ============================ This MP is a variation of SMP4. In the last MP, we built a simulated OS process scheduler. The scheduler can hold only a certain number of processes (workers) at one time. Once the process has been accepted into the scheduler, the scheduler decides in what order the processes execute. We implemented two scheduling algorithms: FIFO and Round Robin. In this MP, we are to simulate a time-sharing system by using signals and timers. We will only implement the Round Robin algorithm. Instead of using iterations to model the concept of "time slices" (as in the last MP), we use interval timers. The scheduler is installed with an interval timer. The timer starts ticking when the scheduler picks a thread to use the CPU which in turn signals the thread when its time slice is finished thus allowing the scheduler to pick another thread and so on. When a thread has completely finished its work it leaves the scheduler to allow a waiting thread to enter. Please note that in this MP, only the timer and scheduler send signals. The threads passively handle the signals without signaling back to the scheduler. The program takes a number of arguments. Arg1 determines the number of jobs (threads in our implementation) created; arg2 specifies the queue size of the scheduler. Arg3 through argN gives the duration (the required time slices to complete a job) of each job. Hence if we create 2 jobs, we should supply arg3 and arg4 for the required duration. You can assume that the autograder will always supply the correct number of arguments and hence you do not have to detect invalid input. Here is an example of program output, once the program is complete: % scheduler 3 2 3 2 3 Main: running 3 workers with queue size 2 for quanta: 3 2 3 Main: detaching worker thread 3075926960. Main: detaching worker thread 3065437104. Main: detaching worker thread 3054947248. Main: waiting for scheduler 3086416816. Scheduler: waiting for workers. Thread 3075926960: in scheduler queue. Thread 3075926960: suspending. Thread 3065437104: in scheduler queue. Thread 3065437104: suspending. Scheduler: scheduling. Scheduler: resuming 3075926960. Thread 3075926960: resuming. Scheduler: suspending 3075926960. Scheduler: scheduling. Scheduler: resuming 3065437104. Thread 3065437104: resuming. Thread 3075926960: suspending. Scheduler: suspending 3065437104. Scheduler: scheduling. Scheduler: resuming 3075926960. Thread 3075926960: resuming. Thread 3065437104: suspending. Scheduler: suspending 3075926960. Scheduler: scheduling. Scheduler: resuming 3065437104. Thread 3065437104: resuming. Thread 3075926960: suspending. Scheduler: suspending 3065437104. Thread 3065437104: leaving scheduler queue. Scheduler: scheduling. Scheduler: resuming 3075926960. Thread 3075926960: resuming. Thread 3065437104: terminating. Thread 3054947248: in scheduler queue. Thread 3054947248: suspending. Scheduler: suspending 3075926960. Thread 3075926960: leaving scheduler queue. Scheduler: scheduling. Scheduler: resuming 3054947248. Thread 3054947248: resuming. Thread 3075926960: terminating. Scheduler: suspending 3054947248. Scheduler: scheduling. Scheduler: resuming 3054947248. Thread 3054947248: suspending. Thread 3054947248: resuming. Scheduler: suspending 3054947248. Scheduler: scheduling. Scheduler: resuming 3054947248. Thread 3054947248: suspending. Thread 3054947248: resuming. Scheduler: suspending 3054947248. Thread 3054947248: leaving scheduler queue. Thread 3054947248: terminating. The total wait time is 12.062254 seconds. The total run time is 7.958618 seconds. The average wait time is 4.020751 seconds. The average run time is 2.652873 seconds. The goal of this MP is to help you understand (1) how signals and timers work, and (2) how to evaluate the performance of your program. You will first implement the time-sharing system using timers and signals. Then, you will evaluate the overall performance of your program by keeping track of how long each thread is idle, running, etc. The program will use these four signals: SIGALRM: sent by the timer to the scheduler, to indicate another time quantum has passed. SIGUSR1: sent by the scheduler to a worker, to tell it to suspend. SIGUSR2: sent by the scheduler to a suspended worker, to tell it to resume. SIGTERM: sent by the scheduler to a worker, to tell it to cancel. You will need to set up the appropriate handlers and masks for these signals. You will use these functions: clock_gettime pthread_sigmask pthread_kill sigaction sigaddset sigemptyset sigwait timer_settime timer_create Also, make sure you understand how the POSIX:TMR interval timer works. There are two ways you can test your code. You can run the built-in grading tests by running "scheduler -test -f0 rr". This runs 5 tests, each of which can be run individually. You can also test you program with specific parameters by running "scheduler -test gen ..." where the ellipsis contains the parameters you would pass to scheduler. Programming =========== Part I: Modify the scheduler code (scheduler.c) ----------------------------------------------- We use the scheduler thread to setup the timer and handle the scheduling for the system. The scheduler handles the SIGALRM events that come from the timer, and sends out signals to the worker threads. Step 1. Modify the code in init_sched_queue() function in scheduler.c to initialize the scheduler with a POSIX:TMR interval timer. Use CLOCK_REALTIME in timer_create(). The timer will be stored in the global variable "timer", which will be started in scheduler_run() (see Step 4 below). Step 2. Implement setup_sig_handlers(). Use sigaction() to install signal handlers for SIGALRM, SIGUSR1, and SIGTERM. SIGALRM should trigger timer_handler(), SIGUSR1 should trigger suspend_thread(), and SIGTERM should trigger cancel_thread(). Notice no handler is installed for SIGUSR2; this signal will be handled differently, in step 8. Step 3. In the scheduler_run() function, start the timer. Use timer_settime(). The time quantum (1 second) is given in scheduler.h. The timer should go off repeatedly at regular intervals defined by the timer quantum. In Round-Robin, whenever the timer goes off, the scheduler suspends the currently running thread, and tells the next thread to resume its operations using signals. These steps are listed in timer_handler(), which is called every time the timer goes off. In this implementation, the timer handler makes use of suspend_worker() and resume_worker() to accomplush these steps. Step 4. Complete the suspend_worker() function. First, update the info->quanta value. This is the number of quanta that remain for this thread to execute. It is initialized to the value passed on the command line, and decreases as the thread executes. If there is any more work for this worker to do, send it a signal to suspend, and update the scheduler queue. Otherwise, cancel the thread. Step 5. Complete the cancel_worker() function by sending the appropriate signal to the thread, telling it to kill itself. Step 6. Complete the resume_worker() function by sending the appropriate signal to the thread, telling it to resume execution. Part II: Modify the worker code (worker.c) ------------------------------------------ In this section, you will modify the worker code to correctly handle the signals from the scheduler that you implemented in the previous section. You need to modify the thread functions so that it immediately suspends the thread, waiting for a resume signal from the scheduler. You will need to use sigwait() to force the thread to suspend itself and wait for a resume signal. You need also to implement a signal handler in worker.c to catch and handle the suspend signals. Step 7. Modify start_worker() to (1) block SIGUSR2 and SIGALRM, and (2) unblock SIGUSR1 and SIGTERM. Step 8. Implement suspend_thread(), the handler for the SIGUSR1 signal. The thread should block until it receives a resume (SIGUSR2) signal. Part III: Modify the evaluation code (scheduler.c) -------------------------------------------------- This program keeps track of run time, and wait time. Each thread saves these two values regarding its own execution in its thread_info_t. Tracking these values requires also knowing the last time the thread suspended or resumed. Therefore, these two values are also kept in thread_info_t. See scheduler.h. In this section, you will implement the functions that calculate run time and wait time. All code that does this will be in scheduler.c. When the program is done, it will collect all these values, and print out the total and average wait time and run time. For your convenience, you are given a function time_difference() to compute the difference between two times in microseconds. Step 9. Modify create_workers() to initialize the various time variables. Step 10. Implement update_run_time(). This is called by suspend_worker(). Step 11. Implement update_wait_time(). This is called by resume_worker(). Questions ========== Question 1. Why do we block SIGUSR2 and SIGALRM in worker.c? Why do we unblock SIGUSR1 and SIGTERM in worker.c? Question 2. We use sigwait() and sigaction() in our code. Explain the difference between the two. (Please explain from the aspect of thread behavior rather than syntax). Question 3. When we use POSIX:TMR interval timer, we are using relative time. What is the alternative? Explain the difference between the two. Question 4. Look at start_worker() in worker.c, a worker thread is executing within an infinite loop at the end. When does a worker thread terminate? Question 5. When does the scheduler finish? Why does it not exit when the scheduler queue is empty? Question 6. After a thread is scheduled to run, is it still in the sched_queue? When is it removed from the head of the queue? When is it removed from the queue completely? Question 7. We've removed all other condition variables in SMP4, and replaced them with a timer and signals. Why do we still use the semaphore queue_sem? Question 8. What's the purpose of the global variable "completed" in scheduler.c? Why do we compare "completed" with thread_count before we wait_for_queue() in next_worker()? Question 9. We only implemented Round Robin in this SMP. If we want to implement a FIFO scheduling algorithm and keep the modification as minimum, which function in scheduler.c is the one that you should modify? Briefly describe how you would modify this function. Question 10. In this implementation, the scheduler only changes threads when the time quantum expires. Briefly explain how you would use an additional signal to allow the scheduler to change threads in the middle of a time quantum. In what situations would this be useful?

__MACOSX/assign3/assign3_part2/._README.txt
assign3/assign3_part2/scheduler.c
#include #include #include #include #include #include #include #include "scheduler.h" #include "worker.h" /* * define the extern global variables here. */ sem_t queue_sem; /* semaphore for scheduler queue */ thread_info_list sched_queue; /* list of current workers */ static int quit = 0; static timer_t timer; static thread_info_t *currentThread= 0; static long wait_times; static long run_times; static int completed = 0; static int thread_count = 0; static void exit_error(int); /* helper function. */ static void wait_for_queue();