Date: Tue, 30 Oct 2012 10:45:41 -0500 From: steven_nikkel@ertyu.org To: freebsd-stable@freebsd.org Subject: Re: CPU Competition Issue Message-ID: <52c0ed4c284d09009a886a09321f477b.squirrel@www2.ertyu.org> In-Reply-To: <508FC3D8.8030800@ShaneWare.Biz> References: <M.O.0.1210292022200.70211@ertyu.org> <508F9671.3060501@FreeBSD.org> <508FC3D8.8030800@ShaneWare.Biz>
next in thread | previous in thread | raw e-mail | index | archive | help
> On 30/10/2012 19:27, Andriy Gapon wrote: >> on 30/10/2012 03:33 Steven Nikkel said the following: >>> I'm running a long duration CPU-centric process that will gobble up all >>> available CPU time. I have it set to run at nice +20. While it's >>> running I've >>> noticed other processes have a hard time getting CPU time and run their >>> activites very slowly. The processes I've noticed issues with are IO >>> involved, >>> but they don't appear to be IO blocked as they run dramatically faster >>> and use >>> much more CPU time when the CPU intensive process is not running. I >>> haven't >>> noticed issues with other processes, but I haven't been looking. If I >>> push my >>> CPU intensive process into idle priority 1, all the other processes >>> return to >>> their normal behaviour as if it's not running. >>> >>> This seems to be a specific behaviour on this one machine running >>> 9.0-RELEASE-p4 >>> on an Atom 330 dual core. I've tried with and without hyperthreading >>> enabled >>> with no noticeable change in behaviour. >> >> Can you try with lower nice value, like +10? >> You want a fix from r228718. AFAIR, it is not in 9.0. >> > > Could it be cache based? The atom's smaller cache causing more cache > misses. > > Would you be running zfs? I am running ZFS. I've tried with doing the I/O on a UFS volume instead of ZFS and the behaviour is the same.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?52c0ed4c284d09009a886a09321f477b.squirrel>