In this chapter, • Thread • Benefits of Threads • Multi-Core Programming • Amdahl law • Parallel Working Types • Thread Libraries • Pthreads topics will be discussed.
2 THREAD • A thread, • Thread ID • Program counter, • A group of register and • Includes a stack structure. • Threads commonly use the program code, data part, operating system resources such as files. • Classic processes have a single thread. • If a process has more than one thread, it can simultaneously do more than one task. • Most of the software applications that work on modern computers are working multithread on today. 3 THREAD (CONTINUES…)
• Applications are developed in the form of a single
process with numerous threads. • A web browser, • can transfer data with a thread, • can display data on the screen with another thread. • A word processing application (Microsoft Word), • can fetch data from the keyboard with a thread, • can perform spell check with a thread, • can edit the screen with another thread. • Most operating systems are Kernel Multi-Threaded. 4 THREAD (CONTINUES…)
• Each thread has its own components without sharing.
Single and Multi-Threaded Process:
code data files code data files
registers stack registers registers registers
stack stack stack
thread thread
single-threaded process multithreaded process
5 MULTI-THREAD SERVER ARCHITECTURE • If a web server process works multithreaded, a separate thread is created for each incoming request and the process continues to listen to the port. • It takes time and costly to create a separate process for each request. (2) create new (1) request thread to service the request client server thread
(3) resume listening
for additional client requests 6 BENEFITS OF THREADS • We can collect the benefits of the use of thread in 4 main headings: 1. Responsiveness: In interactive applications, the user continues to work, even if some blocked, locked or long -term operation is carried out, the interaction with the user continues to work. The system’s responsiveness feature is increased. (Server Architecture) 2. Resource sharing: Processes can share their resources through Shared Memory or Message Passing techniques. Threads can share the memory area and other resources of the process they belong to. 7 BENEFITS OF THREADS (CONTINUES…)
3. Economy: When creating a process, memory and
resource allocation is a high cost. However, as threads share the resources of the process they belong to, the context switch task is executed at a lower cost.(In the Solaris operating system, thread creation is 30 times faster and the context switch task in threads is 5 times faster.) 4. Scalibility: In multi-processing architectures, threads can work simultaneously on different core. However, the process with a single thread structure can only work on a processor. Scalability is the measure of a system's ability to increase or decrease in performance and cost in response to changes in application and system processing demands.
8 MULTI-CORE PROGRAMMING
• For an application with 4 threads running on a core,
concurrent operation means running the threads at certain intervals. • Concurrent operation in multi-core systems means parallel operation of threads by assigning a thread to each core. • Parallelism refers to performing multiple tasks simultaneously. • Concurrency allows multiple tasks to progress together by switching between them in short intervals.
• CPU schedulers • Ensure progress (continuity) by constantly switching between processes and • give the impression of parallelism. • In fact, these processes • They work concurrently. • They do not work in parallel.
11 AMDAHL LAW • Amdahl rule: The performance increase in a system according to the number of cores is expressed as follows: • S: refers to the ratio of the part that must be serially operated in the application (non-parallel), and N: refers to the number of cores.
• In an application, if 75% runs parallel and 25% serial (S=0.25),
• When this application is executed on a system with 2 cores (N=2), the speed increases by 1.6 times. • When the number of cores is 4, a 2.28 times speed increase is achieved. • As the number of cores goes to infinity, the speed increases towards (1/S).
12 TYPES OF PARALLEL WORKING • Basically there are two different types of parallel operation. • Data parallelism • Task parallelism
1. Data Parallelism: It focuses on distributing the
parts of the same data set to cores and executing the same type of operations simultaneously. • For example: If 2 cores are to be used to sum an array with N elements, [0]..[(N/2)-1] element is collected in the 1st core, [N/2]..[N-1] element is collected in the 2nd core.
13 TYPES OF PARALLEL WORKING ( C O N T I N U E S … ) 2. Task Parallelism: It focuses on distributing tasks (threads) to cores. • Each thread performs a separate operation. Different threads can work on the same data or different data. • For example: Threads that perform different statistical calculations on the same array elements using the same data but running on different cores. • Generally, the most applications use hybrid types of both data and task parallel operation.
14 THREAD LIBRARIES
• The Thread Library provides API to create and manage the
programmer. • Two different strategies are used to create multiple thread: • Asynchronous Threading: Parent continues to work simultaneously when it creates a new child thread. Web Server Architecture… • Synchronous Threading: Parent stops its work when child thread creates and continues to work when all child threads end (Fork-Join Strategy). • Asynchronous threading: • When data sharing is low among threads, • Synchronous threading: • Threads are used when data sharing is high.
15 THREAD LIBRARIES ( C O N T I N U E S … )
• Today, three basic thread libraries are used:
1. POSIX Pthreads 2. Windows threads 3. Java threads
16 PTHREADS
• Pthreads is the API defined by the IEEE1003.1c
standard for creating and managing threads. • Linux, Unix, Mac OS X and Solaris operating systems use the Pthreads standard. • Windows does not support the Pthreads standard. • In the Pthreads standard, all threads are created as separate functions. • All threads share the data defined in the global scope.
17 PTHREADS EXAMPLE 1
18 PTHREADS EXAMPLE 1 (CONTINUES…)
19 PTHREADS EXAMPLE 1 DETAILED DESCRIPTION PTHREADS EXAMPLE 2
• In the previous example, a thread was created. Multiple
threads can be created as in the example below.
21 .NET THREADS EXAMPLE 1
22 .NET THREADS EXAMPLE 1
23 PTHREADS EXAMPLE 1 (CONTINUES…)
• compile o gcc -pthread -o SumThread SumThread.c • execute o ./SumThread 1000
24 PTHREADS EXAMPLE 1 (CONTINUES…)
• Let's change the function as follows. Let's add
sleep().
• gcc -pthread -o SumThread SumThread.c
• ./SumThread 1000 • Let's open the second terminal tab and run pstree -p. 25 PTHREADS EXAMPLE 1 (CONTINUES…)
• In Linux, thread gets PID. The process is
treated similarly.
26 PTHREADS EXAMPLE 1 (CONTINUES…)
• What happens if we close the line below?
pthread_join(tid,NULL); • Let's undo the previous change and remove sleep(). • Let's close the pthread_join line. • gcc -pthread -o SumThread SumThread.c • ./SumThread 1000 HAS THE TOTAL RESULT CHANGED? • Then, let's print the sum value to the screen in the thread function. What prints on the screen? • Then add pthread_exit(NULL); it to the last line of the main thread. What does the screen print? 27 PTHREADS EXAMPLE 2
