Arjan van de Ven wrote: >>This is again the usual sched_yield() behavior change that has >>haunted OpenOffice and other applications. > > ... which RH has had since RHL 9 at least. > It's only sort of a behavior change as well though.. and certainly a > mis-assumption on the user of sched_yield()... (the assumption that it > won't be put at the end of the queue is just wrong) OpenLDAP has been finally fixed in 2.3.7. It seems the authors where unaware of the problem, and they wouldn't be able to learn about it without diggin into linux/kernel/sched.c. Neither the Linux man-page nor the POSIX standard give a clue about it. Altough I generally dislike threaded programs and expecially those who spin on locks instead of designing a proper arbitration scheme for shared resources, I consider the new kernel behavior unfair and bogus. It lets low-priority processes steal most CPU time away from high-priority ones. From a system-administrator point of view, it makes nice less functional because it doesn't do what you would expect for processes using specific system calls. -- // Bernardo Innocenti - Develer S.r.l., R&D dept. \X/ http://www.develer.com/ -- fedora-devel-list mailing list fedora-devel-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/fedora-devel-list
