[stlport-bugs] [ stlport-Bugs-1720646 ] Multiprocessor performance problems (Windows 2003)
Brought to you by:
complement
From: SourceForge.net <no...@so...> - 2007-05-18 07:35:22
|
Bugs item #1720646, was opened at 2007-05-17 14:37 Message generated for change (Comment added) made by o-taras You can respond by visiting: https://sourceforge.net/tracker/?func=detail&atid=766244&aid=1720646&group_id=146814 Please note that this message will contain a full copy of the comment thread, including the initial issue submission, for this request, not just the latest update. Category: Operational Environment/Compiler specific code Group: None >Status: Open >Resolution: None >Priority: 5 Private: No Submitted By: Taras Overchuk (o-taras) Assigned to: Nobody/Anonymous (nobody) Summary: Multiprocessor performance problems (Windows 2003) Initial Comment: Currently we faced a problem with performance of STLPort on multiprocessor machine (Windows 2003). After some investigation it was found that the problem is in _STLP_mutex_base implementation for Windows. It uses Sleep(1) to resolve concurrency and this leads to performance problems because: - Sleep(1) call is often reached on multiprocessor platform - Sleep(1) can actually take 15 msec and it is too long As a result multithreaded application uses max 10% of CPU and sleeps most of the time BTW boost doesn't use Sleep(1) for its lightweight mutex any more (original code contains reference to boost.org). OPERATIONAL ENVIRONMENT: - Windows 2003 Server - 4 x Intel Xeon CPU 5150 @ 2.66GHz, 2.00 GB RAM COMPILER VERSION: - VC++ 7.1 (.NET 2003) STLPORT VERSION - 4.6.2 - 5.3.1 RESULT OF UNIT TESTS See attached Tests.zip. It contains: - test sources - possible patch (for 4.6.2 version) - readme.txt file with reproducing instructions and tests results. ---------------------------------------------------------------------- >Comment By: Taras Overchuk (o-taras) Date: 2007-05-18 10:35 Message: Logged In: YES user_id=1794758 Originator: YES Read 5.1.3 Instead of 5.3.1. Are you going to consider the problem or just searching ways to reject the bug without consideration? There is no metter which version. This part of code with Sleep(1) was not changed for ages. Just open _threads.c file and make sure. ---------------------------------------------------------------------- Comment By: Petr Ovtchenkov (complement) Date: 2007-05-17 15:31 Message: Logged In: YES user_id=615813 Originator: NO If you ---------------------------------------------------------------------- Comment By: Taras Overchuk (o-taras) Date: 2007-05-17 14:54 Message: Logged In: YES user_id=1794758 Originator: YES Read 5.1.3 Instead of 5.3.1. Are you going to consider the problem or just searching ways to reject the bug without consideration? There is no metter which version. This part of code with Sleep(1) was not changed for ages. Just open _threads.c file and make sure. ---------------------------------------------------------------------- You can respond by visiting: https://sourceforge.net/tracker/?func=detail&atid=766244&aid=1720646&group_id=146814 |