zentara schrieb: >> I wrote a small server, which accepts connections on a tcp socket and spawns a >> new thread for each connection, which terminates when the socket has been >> closed. >> >> The server runs fine, however, while stresstesting, I experienced a runaway of >> the memory consumption. >> After a few hundred connects and disconnects I'll get an Out of memory. :-( >> >> In order to test threads in perl, I wrote a new script which just spawns and >> terminates threads. ( Attached it below ) >> >> I already figured out that global variables seem to lead to memory leaks, >> however, although the memory size raises slower now, it still raises. ( which >> I didn't expect ). >> > Hi, sorry to take so long to answer, but I kept getting discouraged > by the complexity of your code..... alot of interacting while() loops. > > Sorry for the testcase, today I was confused by the code myself..
> I was just in a discussion with BrowserUk on http://perlmonks.org > about the need to watch for memory gains in threads. > He is adamant that it can be done without reusing threads. > > My point was "sometimes for some weird reason" you need to reuse > threads instead of making them "throwaways". > > Your conclusion about creating a pool at startup is how I would do it. > Although, I would make a pool of threads, instead of preforked. > Create the threads, and put them into a sleep loop, then used shared > variables to wake them as needed. As they are done with their task, put > them back onto "ready queue" and put them back to sleep. > > Your memory use should stay constant this way. > > I would ask your question on http://perlmonks.org and see what they say. > Thanks for your encouragement, mainly I was in doubt whether my test code has some bugs or not. I searched cpan for existing modules which manage a thread pool, but didn't like what I found. Mainly because the code to run by the threads is handled in an eval expression. I did some profiling and testing, if I hand over just a function's name to be run that's about at least 10 times faster. I'd guess the difference will grow with the script's size. I dived into it and wrote some code which manages a pool of threads and can also store the results. There are two more questions. If I got it right it's not possible to make references shared, especially references to functions / modules..? So I'm doing something like the pseudo code below in order to hand over the code to be runned in the threads. ----- sub function{ do something } sub dosomework{ my $f = shift; &$f( @_ ); } &dosomework( "function", @arguments ); ----- Although something like &dosomework( "modulename::function", @args ) is possible, I wasn't able to figure out how to submit objects to an other thread, I guess it's not possible ? Since I did the work, I'd possibly like to publish it, I believe it could be useful to others. Is there a recommended way / would you recommend it at all ? Michael -- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] http://learn.perl.org/
