Chapter 4: Threads & Concurrency: Silberschatz, Galvin and Gagne ©2018 Operating System Concepts

Chapter 4: Threads &
Concurrency
Operating System Concepts Silberschatz, Galvin and Gagne ©2018

Chapter 4: Threads
q Overview
q Multicore Programming
q Multithreading Models
q Thread Libraries
q Implicit Threading
q Threading Issues
q Operating System Examples
Operating System Concepts 2 Silberschatz, Galvin and Gagne ©2018

Objectives
q Identify the basic components of a thread, and contrast threads and

processes
q Describe the benefits and challenges of designing multithreaded

applications
q Illustrate different approaches to implicit threading including thread

pools, fork-join, and Grand Central Dispatch
q Describe how the Windows and Linux operating systems represent

threads
q Design multithreaded applications using the Pthreads, Java, and

Windows threading APIs

Motivation
q Most modern applications are multithreaded
q Threads run within application
q Multiple tasks with the application can be implemented by separate

threads
o Update display
o Fetch data
o Spell checking
o Answer a network request
q Process creation is heavy-weight while thread creation is light-weight
q Can simplify code, increase efficiency
q Kernels are generally multithreaded

Single and Multithreaded Processes

Multithreaded Server Architecture

Benefits
q Responsiveness – may allow continued execution if part of process is

blocked, especially important for user interfaces
q Resource Sharing – threads share resources of process, easier than

shared memory or message passing (IPC)
q Economy – cheaper than process creation, thread switching lower

overhead than context switching
q Scalability – process can take advantage of multicore architectures

Multicore Programming
q Multicore or multiprocessor systems putting pressure on

programmers, challenges include:
o Dividing activities
o Balance
o Data splitting
o Data dependency
o Testing and debugging
q Parallelism implies a system can perform more than one task
simultaneously
q Concurrency supports more than one task making progress
o Single processor / core, scheduler providing concurrency

Concurrency vs. Parallelism
q Concurrent execution on single-core system:
q Parallelism on a multi-core system:

Multicore Programming
q Types of parallelism
o Data parallelism – distributes subsets of the same data across multiple

cores, same operation on each
o Task parallelism – distributing threads across cores, each thread

performing unique operation

Data and Task Parallelism

Amdahl’s Law
q Identifies performance gains from adding additional cores to an

application that has both serial and parallel components
o S is serial portion
o N processing cores
o That is, if application is 75% parallel / 25% serial, moving from 1 to 2

cores results in speedup of 1.6 times
o As N approaches infinity, speedup approaches 1/S
q Serial portion of an application has disproportionate effect on

performance gained by adding additional cores
q But does the law take into account contemporary multicore systems?

Amdahl’s Law

User Threads and Kernel Threads
q User threads - management done by user-level threads library
q Three primary thread libraries:
o POSIX Pthreads
o Windows threads
o Java threads
q Kernel threads - supported by the Kernel
q Examples – virtually all general purpose operating systems, including:

o Windows, Linux, Mac OS X
o iOS, Android

User and Kernel Threads

Multithreading Models
q Many-to-One
q One-to-One
q Many-to-Many

Many-to-One
q Many user-level threads mapped to single kernel thread
q One thread blocking causes all to block
q Multiple threads may not run in parallel on multicore system because

only one may be in kernel at a time
q Few systems currently use this model
q Examples:
o Solaris Green Threads
o GNU Portable Threads

One-to-One
q Each user-level thread maps to one kernel thread
q Creating a user-level thread creates a kernel thread
q More concurrency than many-to-one
q Number of threads per process sometimes restricted due to overhead
q Examples
o Windows
o Linux

Many-to-Many Model
q Allows many user level threads to be mapped to many kernel threads
q Allows the operating system to create a sufficient number of kernel

threads
q Windows with the ThreadFiber package
q Otherwise not very common

Two-level Model
q Similar to Many-to-Many, except that it allows a user thread to be

bound to kernel thread

Thread Libraries
q Thread library provides programmer with API for creating and

managing threads
q Two primary ways of implementing
o Library entirely in user space
o Kernel-level library supported by the OS

Pthreads
q May be provided either as user-level or kernel-level
q A POSIX standard (IEEE 1003.1c) API for thread creation and

synchronization
q Specification, not implementation
q API specifies behavior of the thread library, implementation is up to

development of the library
q Common in UNIX operating systems (Linux & Mac OS X)

Pthreads Example

Pthreads Example (cont)

Pthreads CodeCode
Pthreads for Joining 10 Threads
for Joining 10 Threads
Operating System Concepts ñ 9th Edition 4.21 Silberschatz, Galvin and Gagne ©2013
Windows Multithreaded C Program

Windows Multithreaded C Program (Cont.)

Java Threads
q Java threads are managed by the JVM
q Typically implemented using the threads model provided by

underlying OS
q Java threads may be created by:
o Extending Thread class
o Implementing the Runnable interface
o Standard practice is to implement Runnable interface

Java Threads
q Implementing Runnable interface:
q Creating a thread:
q Waiting on a thread:

Java Executor Framework
q Rather than explicitly creating threads, Java also allows thread

creation around the Executor interface:
q The Executor is used as follows:

Java Executor Framework

Java Executor Framework (cont.)

Implicit Threading
q Growing in popularity as numbers of threads increase, program

correctness more difficult with explicit threads
q Creation and management of threads done by compilers and run-time
libraries rather than programmers
q Five methods explored
o Thread Pools
o Fork-Join
o OpenMP
o Grand Central Dispatch
o Intel Threading Building Blocks

Thread Pools
q Create a number of threads in a pool where they await work
q Advantages:
o Usually slightly faster to service a request with an existing thread than
create a new thread
o Allows the number of threads in the application(s) to be bound to the size
of the pool
o Separating task to be performed from mechanics of creating task allows
different strategies for running task
4 i.e., Tasks could be scheduled to run periodically
q Windows API supports thread pools:

Java Thread Pools
q Three factory methods for creating thread pools in Executors class:

Java Thread Pools (cont)

Fork-Join Parallelism
q Multiple threads (tasks) are forked, and then joined.

q General algorithm for fork-join strategy:


Fork-Join Parallelism in Java


q The ForkJoinTask is an abstract base class
q RecursiveTask and RecursiveAction classes extend

ForkJoinTask
q RecursiveTask returns a result (via the return value from the

compute() method)
q RecursiveAction does not return a result

OpenMP
q Set of compiler directives

and an API for C, C++,
FORTRAN
q Provides support for parallel

programming in shared-
memory environments
q Identifies parallel regions –
blocks of code that can run
in parallel
q #pragma omp parallel
q Create as many threads as

there are cores

OpenMP Example
q Run the for loop in parallel

Grand Central Dispatch
q Apple technology for macOS and iOS operating systems
q Extensions to C, C++ and Objective-C languages, API, and run-time

library
q Allows identification of parallel sections
q Manages most of the details of threading
q Block is in “^{ }” :
ˆ{ printf("I am a block"); }
q Blocks placed in dispatch queue
o Assigned to available thread in thread pool when removed from queue

q Two types of dispatch queues:
o serial – blocks removed in FIFO order, queue is per process, called main
queue
4 Programmers can create additional serial queues within program
o concurrent – removed in FIFO order but several may be removed at a

time
4 Four system wide queues divided by quality of service:
– QOS_CLASS_USER_INTERACTIVE
– QOS_CLASS_USER_INITIATED
– QOS_CLASS_USER_UTILITY
– QOS_CLASS_USER_BACKGROUND

q For the Swift language a task is defined as a closure – similar to a

block, minus the caret
q Closures are submitted to the queue using the dispatch_async()
function:

Intel Threading Building Blocks (TBB)
q Template library for designing parallel C++ programs
q A serial version of a simple for loop
q The same for loop written using TBB with parallel_for

statement:

Threading Issues
q Semantics of fork() and exec() system calls
q Signal handling
o Synchronous and asynchronous
q Thread cancellation of target thread
o Asynchronous or deferred
q Thread-local storage
q Scheduler Activations

Semantics of fork() and exec()
q Does fork() duplicate only the calling thread or all threads?
o Some UNIXes have two versions of fork
q exec() usually works as normal – replace the running process

including all threads

Signal Handling
q Signals are used in UNIX systems to notify a process that a

particular event has occurred.
q A signal handler is used to process signals

o Signal is generated by particular event
o Signal is delivered to a process
o Signal is handled by one of two signal handlers
4 default
4 user-defined
q Every signal has a default handler that kernel runs when handling
signal
o User-defined signal handler can override default
o For single-threaded, signal delivered to process

Signal Handling (Cont.)
q Where should a signal be delivered for multi-threaded?
o Deliver the signal to the thread to which the signal applies
o Deliver the signal to every thread in the process
o Deliver the signal to certain threads in the process
o Assign a specific thread to receive all signals for the process

Thread Cancellation
q Terminating a thread before it has finished
q Thread to be canceled is target thread
q Two general approaches:
o Asynchronous cancellation
terminates the target thread
immediately
o Deferred cancellation allows the

target thread to periodically
check if it should be cancelled
q Pthread code to create and

cancel a thread:
Thread Cancellation (Cont.)
q Invoking thread cancellation requests cancellation, but actual

cancellation depends on thread state
q If thread has cancellation disabled, cancellation remains pending until

thread enables it
q Default type is deferred

o Cancellation only occurs when thread reaches cancellation point
4 i.e., pthread_testcancel()
4 Then cleanup handler is invoked
q On Linux systems, thread cancellation is handled through signals

Thread Cancellation in Java
q Deferred cancellation uses the interrupt() method, which sets the

interrupted status of a thread.
q A thread can then check to see if it has been interrupted:

Thread-Local Storage
q Thread-local storage (TLS) allows each thread to have its own copy
of data
q Useful when you do not have control over the thread creation process
(i.e., when using a thread pool)
q Different from local variables
o Local variables visible only during single function invocation
o TLS visible across function invocations
q Similar to static data
o TLS is unique to each thread

Scheduler Activations
q Both M:M and Two-level models require communication to maintain

the appropriate number of kernel threads allocated to the application
q Typically use an intermediate data structure between user and kernel

threads – lightweight process (LWP)
o Appears to be a virtual processor on which process can schedule user
thread to run
o Each LWP attached to kernel thread
o How many LWPs to create?

q Scheduler activations provide upcalls - a
communication mechanism from the kernel
to the upcall handler in the thread library
q This communication allows an application to

maintain the correct number kernel threads
Operating System Examples
q Windows Threads
q Linux Threads

Windows Threads
q Windows API – primary API for Windows applications
q Implements the one-to-one mapping, kernel-level
q Each thread contains
o A thread id
o Register set representing state of processor
o Separate user and kernel stacks for when thread runs in user mode or
kernel mode
o Private data storage area used by run-time libraries and dynamic link
libraries (DLLs)
q The register set, stacks, and private storage area are known as the
context of the thread

Windows Threads (Cont.)
q The primary data structures of a thread include:
o ETHREAD (executive thread block) – includes pointer to process to

which thread belongs and to KTHREAD, in kernel space
o KTHREAD (kernel thread block) – scheduling and synchronization info,

kernel-mode stack, pointer to TEB, in kernel space
o TEB (thread environment block) – thread id, user-mode stack, thread-

local storage, in user space

Windows Threads Data Structures

Linux Threads
q Linux refers to them as tasks rather than threads
q Thread creation is done through clone() system call
q clone() allows a child task to share the address space of the

parent task (process)
o Flags control behavior
q struct task_struct points to process data structures (shared or

unique)

End of Chapter 4
Operating System Concepts Silberschatz, Galvin and Gagne ©2018

Chapter 4: Threads & Concurrency: Silberschatz, Galvin and Gagne ©2018 Operating System Concepts

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Chapter 4: Threads & Concurrency: Silberschatz, Galvin and Gagne ©2018 Operating System Concepts

Uploaded by

Copyright:

Available Formats

Chapter 4: Threads &

Operating System Concepts Silberschatz, Galvin and Gagne ©2018

q Operating System Examples

Operating System Concepts 2 Silberschatz, Galvin and Gagne ©2018

q Identify the basic components of a thread, and contrast threads and

q Describe the benefits and challenges of designing multithreaded

q Illustrate different approaches to implicit threading including thread

q Describe how the Windows and Linux operating systems represent

q Design multithreaded applications using the Pthreads, Java, and

Operating System Concepts 3 Silberschatz, Galvin and Gagne ©2018

q Most modern applications are multithreaded

q Threads run within application

q Multiple tasks with the application can be implemented by separate

q Can simplify code, increase efficiency

q Kernels are generally multithreaded

Operating System Concepts 4 Silberschatz, Galvin and Gagne ©2018

Operating System Concepts 5 Silberschatz, Galvin and Gagne ©2018

Operating System Concepts 6 Silberschatz, Galvin and Gagne ©2018

q Responsiveness – may allow continued execution if part of process is

q Resource Sharing – threads share resources of process, easier than

q Economy – cheaper than process creation, thread switching lower

q Scalability – process can take advantage of multicore architectures

Operating System Concepts 7 Silberschatz, Galvin and Gagne ©2018

q Multicore or multiprocessor systems putting pressure on

q Concurrency supports more than one task making progress

o Single processor / core, scheduler providing concurrency

Operating System Concepts 8 Silberschatz, Galvin and Gagne ©2018

q Concurrent execution on single-core system:

q Parallelism on a multi-core system:

Operating System Concepts 9 Silberschatz, Galvin and Gagne ©2018

o Data parallelism – distributes subsets of the same data across multiple

o Task parallelism – distributing threads across cores, each thread

Operating System Concepts 10 Silberschatz, Galvin and Gagne ©2018

Operating System Concepts 11 Silberschatz, Galvin and Gagne ©2018

q Identifies performance gains from adding additional cores to an

o That is, if application is 75% parallel / 25% serial, moving from 1 to 2

o As N approaches infinity, speedup approaches 1/S

q Serial portion of an application has disproportionate effect on

Operating System Concepts 12 Silberschatz, Galvin and Gagne ©2018

Operating System Concepts 13 Silberschatz, Galvin and Gagne ©2018

q User threads - management done by user-level threads library

q Three primary thread libraries:

q Kernel threads - supported by the Kernel

q Examples – virtually all general purpose operating systems, including:

Operating System Concepts 14 Silberschatz, Galvin and Gagne ©2018

Operating System Concepts 15 Silberschatz, Galvin and Gagne ©2018

Operating System Concepts 16 Silberschatz, Galvin and Gagne ©2018

q Many user-level threads mapped to single kernel thread

q One thread blocking causes all to block

q Multiple threads may not run in parallel on multicore system because

o Solaris Green Threads

o GNU Portable Threads

Operating System Concepts 17 Silberschatz, Galvin and Gagne ©2018

q Each user-level thread maps to one kernel thread

q Creating a user-level thread creates a kernel thread

q More concurrency than many-to-one

q Number of threads per process sometimes restricted due to overhead

Operating System Concepts 18 Silberschatz, Galvin and Gagne ©2018

q Allows many user level threads to be mapped to many kernel threads

q Allows the operating system to create a sufficient number of kernel

q Otherwise not very common

Operating System Concepts 19 Silberschatz, Galvin and Gagne ©2018

q Similar to Many-to-Many, except that it allows a user thread to be