Threads
CS3008 Operating Systems
Lecture 06
Process
• Process is defined as the unit of resource allocation and
protection
– Protected access to processors, other processes, files, I/O
• Processes embody two characteristic concepts
– Resource ownership:
• Processes own and address space (Virtual memory space to hold the
process image)
• Processes own a set of resources (I/O devices, I/O channels, files, main
memory)
• Operating system protects process to prevent unwanted interference
between processes
– Scheduling / execution
• Process has execution state (running, ready etc.) and dispatching
priority
• Execution follows an execution path (trace) through one or more
programs
• This execution may be interleaved with that of other processes
Threads
• Processes have at least one thread of control
– Is the CPU context, when process is dispatched for
execution
• Goal
– Allow processes to have multiple threads of
control
– Multiple threads run in the same address space,
share the same memory space
• The spawning of a thread only creates a new thread
control structure
Threads
• The unit of dispatching in modern operating systems is
usually a thread
– Represent a single thread of execution within a process
– Operating system can manage multiple threads of
execution within a process
– The thread is provided with its own register context and
stack space
– It is placed in the ready queue
– Threads are also called “lightweight processes”
• The unit of resource ownership is the process
– When a new process is created, at least one thread within
the process is created as well
– Threads share resources owned by a process (code, data,
file handles)
Multithreading
• Multithreading is the ability of an operating
system to support multiple threads of execution
within a single process
• Threads are described by the following:
– Thread execution state
• running, ready, blocked
– Thread Control Block
• A saved thread context when not running (each thread has a
separate program counter)
– An execution stack
– Some per-thread static storage for local variables
– Access to memory and resources of its process, shared
with all other threads of that process
Multithreaded Process Model
Threads
• All threads share the same address space
– Share global variables
• All threads share the same open files, child
processes, signals, etc.
• There is no protection between threads
– As they share the same address space they may
overwrite each others data
• As a process is owned by one user, all threads
are owned by one user
Threads vs Proceses
• Less time to create a new thread (10-times
faster than process creation in Unix)
• Less time to terminate a thread
• Less time to switch between threads
• Threads enhance efficiency in communication
between different communicating programs,
no need to call kernel routines, as all threads
live in same process context
Threads vs Processes
• Advantages of Threads
– Much faster to create a thread than a process
• Spawning a new thread only involves allocating a new stack and a
new thread control block
– Much faster to switch between threads than to switch
between processes
– Threads share data easily
• Disadvantages
– Processes are more flexible
• They don’t have to run on the same processor
– No protection between threads
• Share same memory, may interfere with each other
– If threads are implemented as user threads instead of
kernel threads
• If one thread blocks, all threads in process block
Thread States
Wait-for-event Running 2.
1. Dispatch
Timeout
Blocked 3. Ready
4.
Event-occurred
• Threads have three states
– Running: CPU executes thread
– Ready: thread control block is placed in Ready queue
– Blocked: thread awaits event
Thread Operations
• There are four basic operation for managing threads
– Spawn / create
• A thread is created and provided with its own register context and
stack space, it can spawn further threads
• It is placed on the Ready queue
– Block:
• if a thread waits for an event, it will block and placed on the event
queue
• the processor may switch to another thread in the same or a different
process
– Unblock:
• When the event occurs, for which the thread is waiting, it is placed on
the ready queue
– Finish:
• When a thread completes, its register context and stacks are de-
allocated
Thread Programming
• POSIX standard threads: pthreads
• Describes an API for creating and managing
threads
• There is at least one thread that is created by
executing main()
• Other threads are spawned / created from this
initial thread
POSIX Thread Programming
• Thread creation
pthread_create ( thread, attr, start_routine, arg )
– Returns a new thread ID with parameter “thread”
– Executes the routine specified by “start_routine”
with argument specified by “arg”
• Thread termination
pthread_exit ( status )
– Terminates the thread, sends “status” to any
thread waiting by calling pthread_join()
POSIX Thread Programming
• Thread synchronisation
pthread_join ( threadid, status)
– Blocks the calling thread until the thread specified by
“threadid” terminates
– The argument “status” passes on the return status of
pthread_exit(), called by the thread specified by
“threadid”
• Thread yield
pthread_yield ( )
– Calling thread gives up the CPU and enters the Ready
queue
Thread Programming
Thread Implementation
• Two main categories of thread implementation
– User-level Threads (ULTs)
– Kernel-level Threads (KLTs)
Pure User-Level Pure Kernel-Level CombinedLevel
ULT KLT ULT/KLT
User-Level Threads
• User-Level Threads
– Threads are managed by application, using thread
library functions
– Kernel not aware of the existence of threads, it knows
only processes with one thread of execution
• Benefit
– Light thread switching
• Does not need kernel mode privileges
• Also called “green threads” on some systems (e.g.
Solaris)
User-Level Threads
• Problem
– If one thread blocks, the entire
process is blocked, including all
other threads in it
– Only one thread can access the
kernel at a time, as the process is
the unit of execution known by
kernel
– All threads have to run on the
same processor in a
multiprocessor system
• cannot run in parallel utilising
different processors at the same
time, as process is dispatched on one
processor
Kernel-Level Threads
• Threads are managed by Kernel
– Creating / terminating threads via
system calls
• Benefit
– Fine-grain scheduling done on
thread basis
– If a thread blocks (e.g. waiting for
I/O), another one can be scheduled
by kernel without blocking the
whole process
– Threads can be distributed to
multiple processors and run in
parallel
• Problem
– Heavy thread switching involving
mode switch
Kernel Threads
• Example Systems
– Windows XP/7/8
– Solaris
– Linux
– Mac OSX
Hybrid Implementations
• Try to combine advantages of both user-level and kernel-
level threads
– User-level: light-weight thread switching
– Kernel-level: allows dispatch of other threads (same of different
process), when one threads blocks, true parallelism in
multiprocessor systems possible
• Operating system manages kernel threads and schedules
those for execution
• Mapping of user-level threads onto these kernel threads
• Different Multithreading Models:
– Many-to-one
– One-to-one
– Many-to-many
Many-to-One Model
• All user-level threads of one
process mapped to a single kernel-
level thread
• Thread management in user space
– Efficient
– Application can run its own
scheduler implementation
• One thread can access the kernel at
a time
– Limited concurrency, limited
parallelism
• Examples
– “Green threads” (e.g. Solaris)
– Gnu Portable Threads
One-to-One Model
• Each user-level thread mapped to a kernel thread
• One blocking thread does not block other threads
• Multiple threads access kernel concurrently
• Problem
– Creating a user-level thread requires creation of corresponding
kernel thread
– Kernel may restrict the number of threads created
• Example systems
– Windows, Linux, Solaris 9 (and later) implement the one-to-one
model
Many-to-Many Model
• Many user-level threads are multiplexed (mapped
dynamically) to a smaller or equal number of
kernel threads
– No fixed binding between a user and a kernel
thread
• The number of kernel threads is specific to a
particular application or computer system
– Application may be allocated more kernel threads
on a multiprocessor architecture as on a single-
processor architecture
• No restriction on user-level threads
– Applications can be designed with as many user-
level threads as needed
– Threads are then mapped dynamically onto a
smaller set of currently available kernel threads for
execution
Two-level Model
• Is a variant of the Many-to-Many model, allows a fixed
relationship between a user thread and a kernel thread
• Was used in older Unix-like systems
– IRIX, HP-UX, True64 Unix, Solaris 8
Threading Issues – fork() and exec()
• Semantics of fork() and exec() changes in a
multithreaded program
– Remember:
• fork() creates an identical copy of the calling process
– In case of a multithreaded program
• Should the new process duplicate all threads?
• Or should the new process be created with only one thread?
– If after fork(), the new process calls exec() to start a new program
within the created process image, only one thread may be
sufficient
– Solution: some Unix systems implement two versions
of fork()
Threading Issues – Thread cancellation
• Terminating a thread before it has finished
– this may depend on application
• E.g.: Web browser uses threads to load web page, but when user
selects a link to go to another web page, the loading of the current
web page can be aborted
• Two approaches
– Asynchronous cancellation terminates the target thread
immediately
– Deferred cancellation
• A cancellation request is sent to a target thread, but it continues to
execute undeterred up to a so-called cancellation point
– Particular areas in its program code where it checks for pending cancellation
requests and where a cancellation could take place safely
• If the target thread detects a cancellation request at a cancellation
point, it will end its execution, otherwise it will continue its execution
Threading Issues – Signal Handling
• Signals are used in Unix systems to notify a process
about the occurrence of an event
• All signals follow the same pattern
1. Signal is generated by particular event
2. Signal is delivered to a process
3. Once delivered, a signal must be handled
• In multithreaded systems, there are 4 options
– 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
Threading Issues – Thread Pools
• Threads come with some overhead
• Unlimited thread creation may exhaust memory and CPU
• Solution
– Thread pool: create a number of threads at process startup and put
them in a pool, from where they will be allocated
– When an application needs to spawn a thread, an allocated thread is
taken from the pool and adapted to the application’s needs
• Advantage
– Usually faster to service a request with allready instantiated thread
then creating a new one
– Allows number of threads in applications to be bound by thread pool
size
• Number of pre-allocated threads in pool may depend on
– Number of CPUs, memory size
– Expected number of concurrent requests
