Introduction
I had a requirement to create a server that could run on Windows and Linux platforms. One of the basic requirements of the server was an efficient thread-pooling scheme. I came across the pthreads API. I already had a thread pooling scheme that used IO Completion ports. Using a few defines and pre-processor directives, I changed the existing thread pool to support the pthread library. I have NOT compiled this on Linux yet.
Requirements:
It uses the STL containers vector
and deque
.
It also uses the for_each
STL algorithm.
The classes themselves do not use MFC, but the demo app is created using MFC Single Document App-Wizard.
Requires pthreads library on non-Win32 platforms.
To Use
To use the native Win32 API, comment the line:
#include <pthread/pthread.h&>
in the header file 'defines.h'.
Step 1: Include the header file thread.h.
Step 2: Derive a class CMyJob
from ThreadJob
and implement the 'execute' virtual function. After the job to be done in the thread is complete, do not forget to call the 'execute' function of the base class.
Step 3: In the application's initialization, call
CThread::get_threadpool().start(num);
This creates the threads and waits on them.
Step 4: When you have a job to be queued on the pool,
CMyJob* pjob = new CMyJob
CThread::get_threadpool().add_job(pjob);
The thread pool executes the job and after the job is completed, CMyJob
deletes itself.
Step 5: When the application quits, call
CThread::get_threadpool().stop();
This prevents any more jobs being queued, deletes pending jobs, stops the threads and then returns.
You can obtain the pthreads library from ftp://sources.redhat.com/pub/pthreads-win32/dll-latest to test with pthreads.
In my tests, pthreads on Win32 was a fraction slower than the Win32. This could be because pthreads on Win32 is a wrapper over the native Win32 threads API.
Please send any suggestions to improve this class.