worker thread queue (or thread pool) in C under linux?

[matth45]

I'm porting some code I wrote that uses the Win32 api to linux. Is there a worker thread queue api or library in linux? I'd like to find something similar to the Win32 QueueUserWorkItem, since I'd rather not roll my own thread queue.

There's an intel project under GPL called Threading Building Blocks, but it's a C++ library. I also found libcprops, but the project doesn't seem to be maintained and doesn't actually implement a queue, just a pool. Finally there are kernel workqueues (linux/workqueue.h), but these seem to be made with kernel space driver code in mind. Would they work in a userspace program?

Can anyone point me to a library or api that will let me schedule jobs to be run by a pool of threads with a fixed size?

Thanks!

[paulsm4]

Hi -

It sounds like you might be looking for something a little higher-level, but I'd encourage you to consider pthreads:

http://www.yolinux.com/TUTORIALS/Lin...ixThreads.html
http://www.ibm.com/developerworks/li.../l-posix1.html

It also sounds like you're *not* considering C++ ... but I'd look at "Boost" in preference to Intel's TBB:

http://en.wikipedia.org/wiki/Boost_C%2B%2B_Libraries

IMHO .. PSM

[Seedhom]

Originally Posted by matth45 Can anyone point me to a library or api that will let me schedule jobs to be run by a pool of threads with a fixed size? Thanks!

AFAIK there are no such equivalent set of APIs that are de facto in the linux world. I am not familiar with the projects you referenced, but depending on your needs, implementing pthread pool may not be complex undertaking IMHO. At the other end of the spectrum, there is the open-source resource manager Simple Linux Utility for Resource Management (SLURM) which has a plug-in job scheduler if that is what you are looking for.

[matth45]

Thanks for the links, both of you. I'm somewhat familiar with both pthreads and Boost. My impression was that they would both be decent starting places if I wanted to implement a thread pool myself, but this is exactly what I'm trying to avoid. SLURM seems way too heavy.

The program I'm porting has a section where it runs some functions that traverse a 3D dataset where the columns and planes can alternatively be processed independently. The idea of the thread pool was to amortize the creation cost of threads created to process the rows/cols across many jobs. I assume you guys know that's what I was getting at by your answers, but I suppose a little clarification never hurts. Anyway, I'm looking for speed here, since this happens at video rates, on a single system. (I have a system with many cores, large dataset, etc). I'm no threading expert, so I could see myself easily spending a week creating my own pthreads work queue and debugging it. Do you know any example code for something like this? It seems hard to believe that no one is doing this in C in linux. Or is it easier than I'm thinking?

[matth45]

http://docs.sun.com/app/docs/doc/816-5137/ggedn?a=view

Here's one example.