Hi everybody! Here is my questions:
1) Are there any difference to linux between thread and process?
2) I know that in linux i can determine the thread id with pthread_self() or gettid(). But for the same thread they return different values. What function goes wrong?
3) Can I assign thread to a given CPU on an SMP system?
4) I have multithreaded application. How can i check my self and prove that it works on all CPUs in the system?
You are welcome to discuss.

P.S. Linux laptop 2.6.17-5mdv #1 SMP Wed Sep 13 14:32:31 EDT 2006 i686 Intel(R) Celeron(R) M processor 1400MHz GNU/Linux

I think I can answer 1 & 2 for you.

1) For one, processes are complete separate process. So, to share data, you have to use shared memory, pipes, and other forms of IPC. W/ threads, they share the same adress space, so you don't have to worry about how your going to access the data, just how to do it safely.

2) Pthread's exact implementation is left to the implementer. pthread_self probably won't return a process id, but it could I suppose. If your using pthreads, you should use pthread-related functions, not process related functions. Conceptually, threads and processes could have a n:m relationship, so one single thread may run within multiple processes throughout its lifetime.

according to a certain book on kernel, on linux kernel space, threads and processes are the same.

Can you give me an example of how one single thread may run winthin multiple processes throughout its lifetime?

Thanks, that's what i wanted to see.
Question 1 is closed.

You shouldn't close that question too fast.
The Linux threading model has changed. In 2.4 kernels, the default is the LinuxThreads library, which is not fully conformant with the POSIX standard. Each thread is a process with its own pid.
In kernel 2.6, NPTL (Native Posix Thread Library) is now the default. NPTL uses the CLONE_PID flag so threads share the same pid.

and what about questions 3 and 4?

You may try pthread_setaffinity_np or sched_setaffinity.
Try xcpustate or a similar utility to see if all CPUs are busy.

Strange thing, man pthread_setaffinity_np prints "No manual entry...".


But from manual page of sched_setaffinity i can conclude, this function is linux-specific. I can use gettid() with it to assign given thread to the CPU. So, my code is going to be system-dependent.

Searching google for man pthread_setaffinity_np didn't help to get me specifications. Does anyone have complete description?

Yep, looks like no one contributed a manual page ...
Multi-threading support, implementation and documentation used to be somewhat chaotic with Linux.
It would be too with pthread_setaffinity_np, the _np suffix means "non portable".
As far as I know, pthread_setaffinity_np is a wrapper to sched_setaffinity so you may directly use the latter.

You may also have a look at plpa:
http://www.open-mpi.org/software/plpa/

Can sched_setaffinity() be use to increase the probability of task to be executed right away when the current task is made to sleep? Or the set_nice_priority() is enough?

Moreover, what would be the most efficient way of task creation? pthread_create() or clone()?
Sorry for too many questions, I want to know these too. Thanks!

I'm usure about what you are trying to achieve.
What kind of task ?
Clone is non portable, use fork() if you want to create a process.

By the way, does anybody know, how to evaluate the CPU time, consumed by a given thread?

does this help?

I'm using prstat on Solaris to get thread statistics.
I don't what is the equivalent, if any, with Linux. Perhaps some top option but I'm unsure.