Re: Forking PHP to death!!

Semaphores aren't particularly slow unless you have thousands of  
threads.

The problem with sem_acquire() in PHP is that it is missing the  
sem_op parameter normally found in semop() - sem_op for sem_acquire()  
in PHP is always set to one, which means that it is only useful for

  a) limiting access to a critical resource read-only to a specific  
number of clients; or
  b) full locking of the critical resource to one client at a time,  
read or write.

You can't have both.

For a normal semop() call you can adjust the sem_op parameter so that  
you can have multiple readers / one writer of the critical resource  
in question - the writer will block out the rest of the readers while  
it's writing, but the writer can also be made to wait until after all  
the readers are done (PHP can't do the latter).

I don't know if whoever wrote the sem_* PHP bindings didn't  
understand the semop(2) man page or if it's a limitation of Win32  
semaphores or whatever, but the sem_* functions aren't terribly  
useful in their current state.

-Bob

On Jun 29, 2005, at 3:24 PM, Richard Thomas wrote:

> Anyone on the lists play with pcntl_fork much?
>
> Getting back to my php/flash based chat room and I am playing with  
> making a forked version for performance but having issues figuring  
> out how to smoothly communicate information between all the threads.
>
> Right now I have a msg/shm hybred that works for a simple chat room  
> but doesn't consider constantly changing information that needs to  
> be open to all. Constantly getting and releasing semaphores seems  
> time consuming??
>
>
> --
> Richard Thomas - CEO
> Cyberlot Technologies Group Inc.
> 507.398.4124 - Voice
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: talk-unsubscribe-4zcLI8jJc/rYtjvyW6yDsg@xxxxxxxxxxxxxxxx
>
> Please read and follow the list guidelines:
>  http://www.tcphp.org/mailing_list/guidelines
>
> The tcphp.org mailing list is sponsored by pajunas interactive, inc.


---------------------------------------------------------------------
To unsubscribe, e-mail: talk-unsubscribe-4zcLI8jJc/rYtjvyW6yDsg@xxxxxxxxxxxxxxxx

Please read and follow the list guidelines:
 http://www.tcphp.org/mailing_list/guidelines

The tcphp.org mailing list is sponsored by pajunas interactive, inc.