Re: Low benchmark performance

Simon Derr wrote:
> Hi,
> 
> While running a simple thread wake-up benchmark on a 4-way ia64 system I
> noticed an interesting behaviour.
> 
> The program was stolen from Ian Wienand's pthreadbench suite.
> It is a simple producer/consumer program with 1 producer and N consumers.
> The thing is, with some values of N, the program runs almost 10 times
> slower with NPTL than with LinuxThreads.

Why does the test program use pthread_cond_broadcast() instead of 
pthread_cond_signal() to wake up a consumer?


     /* if the queue is full, signal for worker to clean it */
     if ( condition.value ) {
       pthread_cond_broadcast( &condition.full );
       //pthread_cond_signal( &condition.full );
       pthread_cond_wait( &condition.empty , &condition.mutex);
     }

What is broadcast going to do other than create a burst of scheduling? 
Maybe you just see LinuxThreads deficiencies in that area.

What happens when you use pthread_cond_signal() instead?

Has anybody compared NPTL vs. LinuxThreads performance with Apache 2 
plus worker MPM?  It has a similar model to the test program (but uses 
cond_signal instead of cond_broadcast at steady-state), and with 
LinuxThreads has bad scheduling performance and in particular 
scheculing-after-unlock performance with any interesting number of threads.