Date: Tue, 04 Aug 2015 12:55:47 -0700 From: John Baldwin <jhb@freebsd.org> To: freebsd-hackers@freebsd.org Cc: Dieter BSD <dieterbsd@gmail.com> Subject: Re: Realtime process CPU starvation Message-ID: <3169076.CIxh6P9lj2@ralph.baldwin.cx> In-Reply-To: <CAA3ZYrDXmMXk1ZuP0oGd_6G%2BvKTXtfA6mPAsx_spgojHMnE81Q@mail.gmail.com> References: <CAA3ZYrDXmMXk1ZuP0oGd_6G%2BvKTXtfA6mPAsx_spgojHMnE81Q@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Saturday, June 20, 2015 11:20:38 AM Dieter BSD wrote: > For those who care, I'm running > kern.sched.name: ULE > > My (feeble) understanding is that the scheduler mostly looks at cpu > time, and processes doing i/o actually get a bump *up* in priority, > since because of the way hardware worked in the late 1970s-early 1980s > (PDP-11, VAX, ...) the i/o got useful work done while using very little > cpu time. As a result, jobs that do a lot of i/o can receive more > than their fair share of cpu time. And nice(1) (even rtprio and idprio) > may not have much effect on jobs that are i/o bound. Some form of > ionice is sorely needed. Realtime priorities in 8.<ish> and later are higher than the I/O bump. Only ithreads are higher than realtime in modern versions. -- John Baldwin
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?3169076.CIxh6P9lj2>