Anthony Williams wrote:
"Phil Endecott" <spam_from_boost_dev@chezphil.org> writes:
I have been benchmarking different mutex implementations on Linux. The variations that I have considered are the following:
Thanks for the benchmarks. It may well be worth replacing the pthread_mutex_t-based mutex with a futex-based one in some cases.
I'd just like to encourage this effort. Where I currently work lock performance is a big issue, and no we don't have any small repeatable tests - yet. Most of the tests have been real system tests and do not lend themselves to simple comparisons and also before my time... What I would say is that I will certainly be looking at Phil's code and trying it out. From what I know in other parts of our code base, where performance really counts, we use futexes. So I'd really like to see this in boost::thread even if it is a preprocessor option. Jamie
Anthony