Threads & Concurrency

Course Code: CSC 2209 Course Title: Operating Systems

Dept. of Computer Science

Faculty of Science and Technology

Lecturer No: 05 Week No: 05 Semester: Spring 24-25

Lecturer: NOBORANJAN DEY
Lecture Outline

1. Overview
2. Multicore Programming *
3. Multithreading Models *
4. Thread Libraries
5. Implicit Threading
6. Threading Issues *
7. Operating System Examples
Single and Multithreaded Processes
Multithreaded Server Architecture
Benefits MT (Multi Thread)
 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
 Scalability – process can take advantage of multicore
architectures (utilization of Multiple Processor)
Multicore Programming
 Multicore or multiprocessor systems putting pressure on programmers,
challenges include:
 Dividing activities
 Balance
 Data splitting
 Data dependency
 Testing and debugging
 Parallelism implies a system can perform more than one task simultaneously

 Concurrency supports more than one task making progress

 Single processor / core, scheduler providing concurrency
Concurrency vs. Parallelism
 Concurrent execution on single-core system:

 Parallelism on a multi-core system:

Multicore Programming
 Types of parallelism
 Data parallelism – Data parallelism focuses on distributing subsets of the same
data across multiple computing cores and performing the same operation on each
core.

 Task parallelism – Task parallelism involves distributing not data but tasks
(threads) across multiple computing cores.

 Example: lot of pixels of image or payroll cheques to update

 DP: taking the data and dividing among multiple processors (add only bonus to all..
Divide 50-50 two processor)
 TP: divide the tasks on 2 processor (avg, max, min salary), can use the same or different
data
Data and Task Parallelism
Amdahl’s Law
 Identifies performance gains from adding additional cores to an application that has both
serial and parallel components
 S is serial portion
 1-S parallel portion
 N processing cores

 That is, if application is 75% parallel / 25% serial, moving from 1 to 2 cores results in speedup
of 1.6 times
 As N approaches infinity, speedup approaches 1 / S

Serial portion of an application has disproportionate effect on performance gained by
adding additional cores
 But does the law take into account contemporary multicore systems?
Amdahl’s Law
User Threads and Kernel Threads

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

 Three primary thread libraries:
 POSIX Pthreads
 Windows threads
 Java threads
 Kernel threads - Supported by the Kernel
 Examples – virtually all general purpose operating systems, including:
 Windows
 Linux
 Mac OS X
 iOS
 Android
User and Kernel Threads
Multithreading Models

 Many-to-One

 One-to-One

 Many-to-Many
Many-to-One
 Many user-level threads mapped to
single kernel thread
 One thread blocking causes all to
block
 Multiple threads may not run in
parallel on multicore system because
only one may be in kernel at a time
 Few systems currently use this model
 Examples:
 Solaris Green Threads
 GNU Portable Threads
One-to-One
 Each user-level thread maps to kernel thread

 Creating a user-level thread creates a kernel thread

 More concurrency than many-to-one

 Number of threads per process sometimes

restricted due to overhead
 Examples
 Windows
 Linux
Many-to-Many Model
 Allows many user level threads to be mapped to
many kernel threads
 Allows the operating system to create a sufficient
number of kernel threads
 Windows with the ThreadFiber package

 Otherwise not very common

Two-level Model
 Similar to M:M, except that it allows a user thread
(special root user) to be bound to kernel thread
Implicit Threading
 Growing in popularity as numbers of threads increase, program
correctness more difficult with explicit threads
 Creation and management of threads done by compilers and run-time
libraries rather than programmers
 Five methods explored
 Thread Pools
 Fork-Join
 OpenMP
 Grand Central Dispatch
 Intel Threading Building Blocks
Thread Pools
 Create a number of threads in a pool where they await work
 Advantages:
 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
 Separating task to be performed from mechanics of creating task
allows different strategies for running task
 i.e.Tasks could be scheduled to run periodically
 Windows API supports thread pools:
Java Thread Pools
 Three factory methods for creating thread pools in
Executors class:
Fork-Join Parallelism
 Multiple threads (tasks) are forked, and then joined.
Fork-Join Parallelism
Grand Central Dispatch
 For the Swift language a task is defined as a closure – similar to
a block, minus the caret
 Closures are submitted to the queue using the
dispatch_async() function:
Threading Issues
 Semantics of fork() and exec() system calls

 Signal handling
 Synchronous and asynchronous

 Thread cancellation of target thread

 Asynchronous or deferred

 Thread-local storage

 Scheduler Activations
Semantics of fork() and exec()
 Does fork()duplicate only the calling thread or all threads?
 Some UNIXes have two versions of fork
 Why and when to use
 Exec() immediately after Fork(), then duplication of all process is not
needed
 No Exec() after Fork(), then duplication of all process is needed

 exec() usually works as normal – replace the running

process including all threads
Signal Handling
 Signals are used in UNIX systems to notify a process that a particular
event has occurred.
 A signal handler is used to process signals
 Signal is generated by particular event
 Signal is delivered to a process
 Signal is handled by one of two signal handlers:
 default
 user-defined

 Every signal has default handler that kernel runs when handling signal
 User-defined signal handler can override default
 For single-threaded, signal delivered to process
Signal Handling (cont’d)

 Where should a signal be delivered for multi-threaded?

 Deliver the signal to the thread to which the signal applies
 Deliver the signal to every thread in the process
 Deliver the signal to certain threads in the process
 Assign a specific thread to receive all signals for the process
Thread Cancellation
 Terminating a thread before it has finished
 Thread to be canceled is target thread
 Two general approaches:
 Asynchronous cancellation terminates the target thread
immediately
 Deferred cancellation allows the target thread to periodically
check if it should be cancelled
 Pthread code to create and cancel a thread:
Thread Cancellation (cont’d)

 Invoking thread cancellation requests cancellation, but actual cancellation depends

on thread state

 If thread has cancellation disabled, cancellation remains pending until thread

enables it
 Default type is deferred
 Cancellation only occurs when thread reaches cancellation point
 I.e. pthread_testcancel()
 Then cleanup handler is invoked

 On Linux systems, thread cancellation is handled through signals

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

 Similar to static data

 TLS is unique to each thread
Scheduler Activations
 Both M:M and Two-level models require communication to
maintain the appropriate number of kernel threads allocated to
the application
 Typically use an intermediate data structure between user and
kernel threads – lightweight process (LWP)
 Appears to be a virtual processor on which process can schedule
user thread to run
 Each LWP attached to kernel thread
 How many LWPs to create?
 Scheduler activations provide upcalls - a communication
mechanism from the kernel to the upcall handler in the thread
library
 This communication allows an application to maintain the correct
number kernel threads
Operating System Examples
 Windows Threads

 Linux Threads
Windows Threads
 Windows API – primary API for Windows applications

 Implements the one-to-one mapping, kernel-level

 Each thread contains

 A thread id
 Register set representing state of processor
 Separate user and kernel stacks for when thread runs in user mode or kernel
mode
 Private data storage area used by run-time libraries and dynamic link libraries
(DLLs)
 The register set, stacks, and private storage area are known as the context of
the thread
Windows Threads (cont’d)

 The primary data structures of a thread include:

 ETHREAD (executive thread block) – includes pointer to process to
which thread belongs and to KTHREAD, in kernel space
 KTHREAD (kernel thread block) – scheduling and synchronization
info, kernel-mode stack, pointer to TEB, in kernel space
 TEB (thread environment block) – thread id, user-mode stack,
thread-local storage, in user space
Windows Threads Data Structures
Linux Threads
 Linux refers to them as tasks rather than threads
 Thread creation is done through clone() system call
 clone() allows a child task to share the address space of the parent task
(process)
 Flags control behavior

 struct task_struct points to process data structures (shared or unique)

Books
 Operating Systems Concept
 Written by Galvin and Silberschatz
 Edition: 9th
References
 Operating Systems Concept
 Written by Galvin and Silberschatz
 Edition: 9th