On Dec 3 21:36, Takashi Yano wrote: > On Mon, 2 Dec 2024 15:25:54 +0100 > Corinna Vinschen wrote: > > On Nov 29 20:58, Takashi Yano wrote: > > > On Wed, 27 Nov 2024 17:53:53 +0100 > > > Corinna Vinschen wrote: > > > Hmmm, just setting THREAD_PRIORITY_NORMAL might be appropriate. > > > See v3 patch. > > > > > > > The culprit of the behaviour you're seeing is the fact that *all* > > > > cygthread's are running with THREAD_PRIORITY_HIGHEST prio. > > > > > > > > Maybe it's time to rethink this. Most (none?) of the cygthreads really > > > > need highest priority. This *may* have been useful when we only had a > > > > single CPU core, but these times have gone by, and cygthreads serve > > > > quite a few tasks which don't need THREAD_PRIORITY_HIGHEST. > > > > > > > > We could try to start all threads with normal priority, and > > > > only threads suffering from priority problems could be moved to > > > > another prio. > > > > > > Enough testing will be necessary for that, I think. > > > > I see what you mean, so yeah, let's try it your way and cherry-pick > > into 3.5. > > You mean applying the 4th patch to master branch, then apply another > patch which drops setting THREAD_PRIORITY_HIGHEST. Right?
Yes, that's what I meant. Sorry if that wasn't quite clear. > > For the main branch, we should really try to drop setting all > > cygthreads to THREAD_PRIORITY_HIGHEST and leave it as the discretion > > of the thread itself to manage its priority. > > > > Also, even if a higher prio is required for one thread or another, > > THREAD_PRIORITY_ABOVE_NORMAL might be sufficient in most cases. > > Just in my short trial, there is no problem even without HIGHEST > proiority. I'd test that more. Great, thanks! Corinna
