Threads
Last updated
Last updated
In a traditional operating system, a process includes:
an address space (code and data of the associated program).
a set of communication channels with I/O devices.
a single thread of control, which incorporates the processor registers (including the program counter) and a stack.
However, these components can be managed separately.
In this model, thread appears as an execution component within a process.
Several independent threads can coexist in the same process, thus sharing the same address space and the same I/O context.
This is referred to as multithreading.
Threads can be seen as light weight processes.
Each thread is typically associated to the execution of a function that implements some specific activity.
Communication between threads can be done through the process data structure, which is global from the threads point of view.
It includes static and dynamic variables (heap memory).
The main program, also represented by a function that implements a specific activity, is the first thread to be created and, in general, the last to be destroyed.
Threads are implemented by a library, at user level, which provides creation and management of threads without kernel intervention.
versatile and portable.
when a thread calls a blocking system call, the whole process blocks.
because the kernel only sees the process.
Threads are implemented directly at kernel level.
less versatile and less portable.
when a thread calls a blocking system call, another thread can be schedule to execution.
Easier implementation of applications - in many applications, decomposing the solution into a number of parallel activities makes the programming model simpler.
since the address space and the I/O context is shared among all threads, multithreading favors this decomposition.
Better management of computer resources – creating, destroying and switching threads is easier then doing the same with processes.
Better performance – when an application involves substantial I/O, multithreading allows activities to overlap, thus speeding up its execution.
Multiprocessing – real parallelism is possible if multiples CPUs exist.
