Thread Pools

As a quick recap, note that the linux kernel creates an object of the data structure called a "task". Both of what we call as processes and threads are really tasks that are "cloned" from this original task. fork() is one kind of cloning and kthread_create (pthread_create, (portable kernel thread create) API) is another kind of cloning. Multiple threads may share the virtual address space (and thus in turn Data sections, etc), but threads mostly have their own stacks.

A task is a unit of scheduling on by the linux scheduler.

Note that "process" is a task that has a "user level context". When "user level threads" are created, they do not call the clone system call and thus share the address space of the process. These kinds of user-level-threads are called lightweight processes.

One can, if one wants, instead of creating new a thread every time the client asks the server to perform some task, we can have a pool of threads -- what this means is that we can store the "context" (i.e. registers etc) in some place in RAM and toggle between these places with a pointer i.e. all the operating system needs to do to perform a thread switch, is to change the thread pointer (see this). This is essentially an optimization to save the overhead of creating and deleting threads.

Have a look at this link for a quick high-level intro of linux memory management.