Date: Mon, 17 Sep 2007 14:22:29 -0700 (PDT) From: Jeff Roberson <jroberson@chesapeake.net> To: Roman Bogorodskiy <novel@freebsd.org> Cc: freebsd-current@freebsd.org Subject: Re: SCHED_ULE on desktop system Message-ID: <20070917141820.M558@10.0.0.1> In-Reply-To: <20070917162400.GA42669@underworld.novel.ru> References: <20070916061932.GA93480@underworld.novel.ru> <20070915234855.G531@10.0.0.1> <20070916071421.GA1320@underworld.novel.ru> <20070916003323.U4507@10.0.0.1> <20070917044351.GA17565@underworld.novel.ru> <20070917162400.GA42669@underworld.novel.ru>
next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, 17 Sep 2007, Roman Bogorodskiy wrote: > Roman Bogorodskiy wrote: > >> Jeff Roberson wrote: >> >>> On Sun, 16 Sep 2007, Roman Bogorodskiy wrote: >>> >>>> Jeff Roberson wrote: >>>> >>>>> This is contrary to the experiences of many others. Can you send me your >>>>> dmesg? There may be something about your particular hardware that is >>>>> triggering a bug. ULE is definitely designed to be responsive on the >>>>> desktop. >>>> >>>> Thanks for the quick answer! >>>> >>>> Here's my dmesg: http://people.freebsd.org/~novel/misc/dmesg.txt >>> >>> Roman, >>> >>> The enclosed patch helps things on my system, however, there are still some >>> delays due to IO issues. Let me know if this helps. >> >> This patch seems to make my system react faster. However I need some >> time to test it more carefully. > > It looks like it's not really so, it's still very slow under load. :( Roman, Can you please verify that you are not swapping. If you are, try the patch that I sent. Otherwise, can you try getting KTR_SCHED output for me? Put the following in your kernel config file: options KTR options KTR_COMPILE=KTR_SCHED options KTR_MASK=KTR_SCHED options KTR_ENTRIES=65536 Then run the load that exhibits the problem. compile, play a movie, etc. wait for a particularly bad pause and then run the following: sysctl debug.ktr.mask=0 && ktrdump -ct > out && sysctl debug.ktr.mask=536870 It's best to have the command ready in a shell so you can run it in the closest proximity to the incident. Then gzip the results and email them to me. On my system I'm not able to play a movie while running buildworld due to long disk waits. However, if I use a dvd driver, mplayer and x windows both get enough cpu time to play the movie skip free. So this is not a cpu scheduler problem as such. IO scheduling does feel less responsive than 6.x but I have not done a side by side comparison. Jeff > >>> Thanks, >>> Jeff >>> >>>> >>>> Roman Bogorodskiy >>>> >> >> >> Roman Bogorodskiy > > > Roman Bogorodskiy >
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20070917141820.M558>