®Occurs when target checks for cancellation signal
±Allows cancellation at “safe” points
®Called cancellation points in Pthreads
Signal Handling
±Signals are used in UNIX to notify a process that a particular event has occurred
±A signal handler is used to process signals
4Signal is generated by a particular event
4Signal is delivered to a process
4Signal is handled (or ignored/blocked)
±Options:
®Deliver the signal to the thread to which the signal applies
®Applicable with synchronous signals e.g., illegal memory access
®Deliver the signal to every thread in the process
®Deliver the signal to certain threads in the process
®Assign a specific threat to receive all signals for the process
Thread Pooling
±When design situations arise that could benefit by using many short-lived threads, thread pooling is a useful technique.
±Rather than create a brand new thread for each task, you can have one of the threads from the thread pool pulled out of the pool and assigned to the task.
±When the thread is finished with task, it adds itself back to the pool and waits for another assignment
Benefits of Thread Pooling
±You can reduce response time because a thread is already constructed and started and is simply waiting for its next task. In the case of an HTTP server, an available thread in the pool can deliver each new file requested
±Threads are fixed in size at the time of construction. All the threads are started, and then each goes into a wait state until a task is assigned to it. If all the threads are currently assigned a task, new service requests will put into a wait state until one of the threads finishes its task and returns itself to the pool.
Thread Pools
±Motivating example: a web server running on an SMP machine
±To handle each connection:
lCreate a new process to handle it
®too slow, inefficient
lCreate a new thread to handle it
±Option 2 better but still has some problems:
®Some overhead for thread creation/deletion
®Thread will only be used for this connection
®Unbounded number of threads might crash web server
±Better solution: use a thread pool of (usually) fixed size
Thread Pools (cont’d)
±Threads in pool sit idle
±Request comes in:
®Wake up a thread in pool
®Assign it the request
®When it completes, return it to pool
®If no threads in pool available, wait
±Advantages:
®Usually slightly faster to wake up an existing thread than create a new one
®Allows the number of threads in the application to be limited by the size of the pool
®More sophisticated systems dynamically adjust pool size
Thread-specific Data
±Allows each thread to have its own copy of data
±Useful for implementing protection
®For example, user connects to bank’s database server
®Server process responds, has access to all accounts
®Multiple people may be accessing their accounts at the same time
®Thread assigned to manipulate or view user’s bank account
®Using thread-specific data limits (unintentional or erroneous) access by other threads
Pthreads
±A POSIX standard (IEEE 1003.1c) API for thread creation and synchronization
®NOT an implementation
®Implementation in different OSes may be using user or kernel threads
±Common in UNIX operating systems (Solaris, Linux, Mac OS X)
±Fairly portable
±Available on pyrite
±Has man pages on Linux (check “man pthread_create”)
±To use:
®#include <pthread.h>
®Link with pthread library: g++ prog.cc -lpthread
Windows XP Threads
±Implements the one-to-one mapping
±Each thread contains
®A thread id
®Register set
®Separate user and kernel stacks
®Private data storage area
±The register set, stacks, and private storage area are known as the context of the threads
Linux Threads
±Linux refers to them as tasks rather than threads
±Thread creation is done through the clone() system call
±clone() allows a child task to share different things with the parent task such as:
®the address space
®the table of open files
®the signal handlers
±If nothing is shared: same as fork(), creates essentially a process
±If everything is shared: creates a thread as we know it
Linux Threads (cont’d)
Java Threads
±Java threads are managed by the JVM
±Java threads may be created by:
®Extending Thread class
®Implementing the Runnable interface
±Details in textbook
Linux Programming Pthread
Thread Basic
±Thread operations include thread creation, termination, synchronization, data management
±Threads in the same process share:
±Process address space, instructions and most data
±Opened files
±Signals and signal handles
±Each thread has a unique:
±Thread ID
±A register set
±Stack
±Local variables and return address
Pthreads: POSIX Threads
±Pthreads is a standard set of C library functions for multithreaded programming
®IEEE Portable Operating System Interface, POSIX, section 1003.1 standard, 1995
±Programs must be linked with the pthread library (-lpthread)
Pthreads Naming Convention
±Types: pthread[_object]_t
±Functions: pthread[_object]_action
±Constants/Macros: PTHREAD_PURPOSE
±Examples:
®pthread_t: the type of a thread
®pthread_create(): creates a thread
®pthread_mutex_t: the type of a mutex lock
®pthread_mutex_lock(): lock a mutex
®PTHREAD_CREATE_DETACHED
Pthread – POSIX thread
±Include pthread header file
±# include <pthread.h>
±Compile and execution
±C compiler: gcc –o foo foo.c -lpthread
±C++ compiler: g++ -o foo foo.c –lpthread
±Execution: ./foo
Example
Pthread – POSIX thread
±Basic pthread functions
®pthread_create – create child thread
®pthread_exit – thread termination
®pthread_join – wait for thread termination
®phtread_mutex_lock – lock critical section
®pthread_mutex_unlock – unlock critical section
®pthread_cond_wait – wait for a condition signal
®phtread_cond_signal – wake up one waiting thread
®pthread_cond_broadcast – wake up all waiting threads
pthread_self()
±Returns the thread identifier for the calling thread
®At any point in its instruction stream a thread can figure out which thread it is
®Convenient to be able to write code that says: “If you’re thread 1 do this, otherwise do that”
®However, the thread identifier is an opaque object (just a pthread_t value)
®you must use pthread_equal() to test equality
pthread_t pthread_self(void);
int pthread_equal(pthread_t id1, pthread_t id2);
pthread_create()
±Creates a new thread
int pthread_create (
pthread_t *thread,
pthread_attr_t *attr,
void * (*start_routine) (void *),
void *arg);
®Returns 0 to indicate success, otherwise returns error code
®thread: output argument for the id of the new thread
®attr: input argument that specifies the attributes of the thread to be created (NULL = default attributes)
®start_routine: function to use as the start of the new thread
®must have prototype: void * foo(void*)
®arg: argument to pass to the new thread routine
®If the thread routine requires multiple arguments, they must be passed bundled up in an array or a structure
pthread_create() example
±Want to create a thread to compute the sum of the elements of an array
void *do_work(void *arg);
±Needs three arguments
®the array, its size, where to store the sum
®we need to bundle them in a structure
struct arguments {
double *array;
int size;
double *sum;
}
pthread_create() example
int main(int argc, char *argv) {
double array[100];
double sum;
pthread_t worker_thread;
struct arguments *arg;
arg = (struct arguments *)calloc(1,
sizeof(struct arguments));
arg->array = array;
arg->size=100;
arg->sum = ∑
if (pthread_create(&worker_thread, NULL,
do_work, (void *)arg)) {
fprintf(stderr,”Error while creating thread/n”);
exit(1);
}
...
}
pthread_create() example
void *do_work(void *arg) {
struct arguments *argument;
int i, size;
double *array;
double *sum;
argument = (struct arguments*)arg;
size = argument->size;
array = argument->array;
sum = argument->sum;
*sum = 0;
for (i=0;i<size;i++)
*sum += array[i];
return NULL;
}
Comments about the example
±The “main thread” continues its normal execution after creating the “child thread”
±IMPORTANT: If the main thread terminates, then all threads are killed!
®We will see that there is a join() function
±Of course, memory is shared by the parent and the child (the array, the location of the sum)
®nothing prevents the parent from doing something to it while the child is still executing
®which may lead to a wrong computation
®we will see that Pthreads provide locking mechanisms
±The bundling and unbundling of arguments is a bit tedious
Memory Management of Args
±The parent thread allocates memory for the arguments
±Warning #1: you don’t want to free that memory before the child thread has a chance to read it
®That would be a race condition
®Better to let the child do the freeing
±Warning #2: if you create multiple threads you want to be careful there is no sharing of arguments, or that the sharing is safe
®For instance, if you reuse the same data structure for all threads and modify its fields before each call to pthread_create(), some threads may not be able to read the arguments destined to them
®Safest way: have a separate arg structure for each thread
pthread_exit()
±Terminates the calling thread
void pthread_exit(void *retval);
®The return value is made available to another thread calling a pthread_join() (see next slide)
®My previous example had the thread just return from function do_work()
®In this case the call to pthread_exit() is implicit
®The return value of the function serves as the argument to the (implicitly called) pthread_exit().
pthread_join()
±Causes the calling thread to wait for another thread to terminate
int pthread_join(
pthread_t thread,
void **value_ptr);
®thread: input parameter, id of the thread to wait on
®value_ptr: output parameter, value given to pthread_exit() by the terminating thread (which happens to always be a void *)
®returns 0 to indicate success, error code otherwise
®multiple simultaneous calls for the same thread are not allowed
pthread_kill()
±Causes the termination of a thread
int pthread_kill(
pthread_t thread,
int sig);
®thread: input parameter, id of the thread to terminate
®sig: signal number
®returns 0 to indicate success, error code otherwise
fprintf(stderr,”Error while waiting for thread/n”);
exit(1);
}
}
pthread_join()Warning
±This is a common “bug” that first-time pthread programmers encounter
±Without the call to pthread_join() the previous program may end immediately, with the main thread reaching the end of main() and exiting, thus killing all other threads perhaps even before they have had a chance to execute
pthread_join()Warning
±When creating multiple threads be careful to store the handle of each thread in a separate variable
®Typically one has an array of thread handles
±That way you’ll be able to call pthread_join() for each thread
±Also, note that the following code is sequential!
for (i=0; i < num_threads; i++) {
pthread_create(&(threads[i]),...)
pthread_join(threads[i],...)
}
Thread Attributes
±One of the parameters to pthread_create() is a thread attribute
±In all our previous examples we have set it to NULL
±But it can be very useful and provides a simple way to set options:
®Initialize an attribute
®Set its value with some Pthread API call
®Pass it to Pthread API functions like pthread_create()
pthread_attr_init()
±Initialized the thread attribute object to the default values
int pthread_attr_init(
pthread_attr_t *attr);
®Return 0 to indicate success, error code otherwise
®attr: pointer to a thread attribute
Detached Thread
±One option when creating a thread is whether it is joinable or detached
®Joinable: another thread can call join on it
®By default a thread is joinable
®Detached: no thread can call join on it
±Let’s look at the function that allows to set the “detached state”
pthread_attr_setdetachstate()
±Sets the detach state attribute
int pthread_attr_setdetachstate(
pthread_attr_t *attr,
int detachstate);
®returns 0 to indicate success, error code otherwise
®attr: input parameter, thread attribute
®detachstate: can be either
®PTHREAD_CREATE_DETACHED
®PTHREAD_CREATE_JOINABLE (default)
Detach State
±Detached threads have all resources freed when they terminate
±Joinable threads have state information about the thread kept even after they finish
®To allow for a thread to join a finished thread
®So-called “no rush to join”
±So, if you know that you will not need to join a thread, create it in a detached state so that you save resources
±This is lean-and-mean C, as opposed to hand-holding Java, and every little saving is important