Threads

Threads
Amir H. Payberah
amir@sics.se
Amirkabir University of Technology
(Tehran Polytechnic)
Amir H. Payberah (Tehran Polytechnic) Threads 1393/7/12 1 / 45

Motivation

Thread
A basic unit of CPU utilization.

Threads (1/3)
A traditional process: has a single thread.

Threads (1/3)
Multiple threads in a process: performing more than one task at a
time.

Threads (1/3)
Multiple threads in a process: performing more than one task at a
time.
Threads in a process share code section, data section, and other OS
resources, e.g., open ﬁles.

Threads (2/3)
Multiple tasks of an application can be implemented by separate
threads.
• Update display
• Fetch data
• Spell checking
• Answer a network request

Threads (3/3)
Multi-threaded web-server architecture

Threads Beneﬁts (1/2)
Responsiveness: may allow continued execution if part of process is
blocked, especially important for user interfaces.

Responsiveness: may allow continued execution if part of process is
blocked, especially important for user interfaces.
Resource Sharing: threads share resources of process, easier than
shared memory or message passing.

Economy: cheaper than process creation, thread switching lower
overhead than context switching.

Economy: cheaper than process creation, thread switching lower
overhead than context switching.
Scalability: process can take advantage of multiprocessor architec-
tures

Context Switching vs. Threads Switching
Inter-process switching: context switching from process-to-process.

Intra-process switching: switching from thread-to-thread.

Intra-process switching: switching from thread-to-thread.
The cost of intra-process switching is less than the cost of inter-
process switching.

Multicore Programming

Multiprocessor and Multicore Systems (1/2)
Users need more computing performance: single-CPU → multi-CPU

Users need more computing performance: single-CPU → multi-CPU
A similar trend in system design: place multiple computing cores on
a single chip.
• Each core appears as a separate processor.
• Multi-core systems.

Multi-threaded programming
• More eﬃcient use of multiple cores.
• Improved concurrency.

Concurrency vs. Parallelism (1/2)
Parallelism
• Performing more than one task simultaneously.

Parallelism
Concurrency
• Supporting more than one task by allowing all the tasks to make
progress.

Parallelism
Concurrency
progress.
• Single processor/core: scheduler providing concurrency.

Parallelism
Concurrency
progress.
• Single processor/core: scheduler providing concurrency.
• It is possible to have concurrency without parallelism.

Concurrent execution on a single-core system.
Parallelism on a multi-core system.

Types of Parallelism
Data parallelism
• Distributes subsets of the same data across multiple cores, same
operation on each.

Types of Parallelism
Data parallelism
• Distributes subsets of the same data across multiple cores, same
operation on each.
Task parallelism
• Distributes threads across cores, each thread performing unique
operation.

Programming Challenges
Dividing activities
Balance
Data splitting
Data dependency
Testing and debugging

Multi-threading Models

User Threads and Kernel Threads
User threads: management done by user-level threads library.
Kernel threads: supported by the Kernel.

• Three primary thread libraries:
• POSIX pthreads
• Windows threads
• Java threads

• Three primary thread libraries:
• POSIX pthreads
• Windows threads
• Java threads
• All general purpose operating systems, including:
Windows, Solaris, Linux, Tru64 UNIX, Mac OS X

Multi-Threading Models
Many-to-One
One-to-One
Many-to-Many

Many-to-One Model
Many user-level threads mapped to single kernel thread.

Many-to-One Model
One thread blocking causes all to block.

Many-to-One Model
Multiple threads may not run in parallel on multicolor system
because only one may be in kernel at a time.

Many-to-One Model
Multiple threads may not run in parallel on multicolor system
because only one may be in kernel at a time.
Few systems currently use this model.
• Solaris green threads
• GNU portable threads

One-to-One Model
Each user-level thread maps to a kernel thread.

One-to-One Model
Creating a user-level thread creates a kernel thread.

One-to-One Model
More concurrency than many-to-one.

One-to-One Model
Number of threads per process sometimes restricted due to
overhead.

One-to-One Model
Number of threads per process sometimes restricted due to
overhead.
Examples:
• Windows
• Linux
• Solaris 9 and later

Many-to-Many Model
Allows many user-level threads to be mapped to many kernel
threads.

Many-to-Many Model
threads.
Allows the OS to create a suﬃcient number of kernel threads.

Many-to-Many Model
threads.
Allows the OS to create a suﬃcient number of kernel threads.
Examples:
• Solaris prior to version 9
• Windows with the ThreadFiber package

Thread Libraries

Thread Libraries (1/2)
Thread library provides programmer with API for creating and
managing threads.

Thread library provides programmer with API for creating and
managing threads.
Two primary ways of implementing:
• Library entirely in user-space.
• Kernel-level library supported by the OS.

Pthread
• Either a user-level or a kernel-level library.

Pthread
Windows thread
• Kernel-level library.

Pthread
Windows thread
• Kernel-level library.
Java thread
• Uses a thread library available on the host system.

Pthreads
A POSIX API for thread creation and synchronization.

Pthreads
Speciﬁcation, not implementation.

Pthreads
API speciﬁes behavior of the thread library, implementation is up
to development of the library.

Pthreads
API speciﬁes behavior of the thread library, implementation is up
to development of the library.
Common in UNIX OSs, e.g., Solaris, Linux, Mac OS X

Thread ID
The thread ID (TID) is the thread analogue to the process ID (PID).

Thread ID
The PID is assigned by the Linux kernel, and TID is assigned in the
Pthread library.

Thread ID
Pthread library.
Represented by pthread t.

Thread ID
Pthread library.
Represented by pthread t.
Obtaining a TID at runtime:
#include <pthread.h>
pthread_t pthread_self(void);

Creating Threads
pthread create() deﬁnes and launches a new thread.
int pthread_create(pthread_t *thread, const pthread_attr_t *attr,
void *(*start_routine)(void *), void *arg);
start routine has the following signature:
void *start_thread(void *arg);

Terminating Threads
Terminating yourself by calling pthread exit().
void pthread_exit(void *retval);
Terminating others by calling pthread cancel().
int pthread_cancel(pthread_t thread);

Joining and Detaching Threads
Joining allows one thread to block while waiting for the termination
of another.
int pthread_join(pthread_t thread, void **retval);
int pthread_detach(pthread_t thread);
[https://computing.llnl.gov/tutorials/pthreads/#Joining]

Joining and Detaching Threads
Joining allows one thread to block while waiting for the termination
of another.
You use join if you care about what value the thread returns when
it is done, and use detach if you do not.
int pthread_join(pthread_t thread, void **retval);
int pthread_detach(pthread_t thread);
[https://computing.llnl.gov/tutorials/pthreads/#Joining]

A Threading Example
void *start_thread(void *message) {
printf("%sn", (const char *)message);
return message;
}
int main(void) {
pthread_t thread11, thread2;
const char *message1 = "Thread 1";
const char *message2 = "Thread 2";
// Create two threads, each with a different message.
pthread_create(&thread1, NULL, start_thread, (void *)message1);
pthread_create(&thread2, NULL, start_thread, (void *)message2);
// Wait for the threads to exit.
pthread_join(thread1, NULL);
pthread_join(thread2, NULL);
return 0;
}

Implicit Threading

Implicit Threading
Increasing the number of threads: program correctness more diﬃcult
with explicit threads.

Implicit Threading
Implicit threading: creation and management of threads done by
compilers and run-time libraries rather than programmers.

Implicit Threading
Implicit threading: creation and management of threads done by
compilers and run-time libraries rather than programmers.
Three methods explored:
• Thread Pools
• OpenMP
• Grand Central Dispatch

Thread Pools
Create a number of threads in a pool where they await work.

Thread Pools
Usually slightly faster to service a request with an existing thread
than create a new thread.

Thread Pools
Usually slightly faster to service a request with an existing thread
than create a new thread.
Allows the number of threads in the application(s) to be bound to
the size of the pool.

OpenMP (1/2)
Set of compiler directives and APIs for C, C++, FORTRAN.

OpenMP (1/2)
Identiﬁes parallel regions: blocks of code that can run in parallel.

OpenMP (1/2)
Identiﬁes parallel regions: blocks of code that can run in parallel.
#pragma omp parallel: create as many threads as there are cores.
#pragma omp parallel for: run for loop in parallel.

OpenMP (2/2)
#include <omp.h>
#include <stdio.h>
int main(int argc, char *argv[]) {
/* sequential code */
#pragma omp parallel
{
printf("I am a parallel region.");
}
/* sequential code */
return 0;
}

Grand Central Dispatch (1/2)
Apple technology for Mac OS X and iOS: extensions to C, C++
API, and run-time library.

Allows identiﬁcation of parallel sections.

Block is in ˆ{ }: ˆ{ printf("I am a block"); }

Block is in ˆ{ }: ˆ{ printf("I am a block"); }
Blocks placed in dispatch queue.

Two types of dispatch queues:
Serial: blocks removed in FIFO order, queue is per process.
Concurrent: removed in FIFO order but several may be removed at
a time.
dispatch_queue_t queue = dispatch_get_global_queue
(DISPATCH QUEUE PRIORITY DEFAULT, 0);
dispatch_async(queue, ^{ printf("I am a block."); });

Threading Issues

Semantics of fork() and exec()
Does fork() duplicate only the calling thread or all threads?
• Some UNIXes have two versions of fork.

Semantics of fork() and exec()
Does fork() duplicate only the calling thread or all threads?
• Some UNIXes have two versions of fork.
exec() usually works as normal: replaces the running process in-
cluding all threads.

Signal Handling
Where should a signal be delivered for multi-threaded?

Signal Handling
• Deliver the signal to the thread to which the signal applies.

Signal Handling
• Deliver the signal to every thread in the process.

Signal Handling
• Deliver the signal to certain threads in the process.

Signal Handling
• Deliver the signal to certain threads in the process.
• Assign a speciﬁc thread to receive all signals for the process.

Summary

Summary
Single-thread vs. Multi-thread

Summary
Interprocess vs. Intraprocess context switchings

Summary
Concurrency vs. parallelism

Summary
Multi-threading models: many-to-one, one-to-one, many-to-many

Summary
Multi-thread libraries: pthread

Summary
Multi-thread libraries: pthread
Implicit threading

Questions?
Acknowledgements
Some slides were derived from Avi Silberschatz slides.

Threads

More Related Content

Threads