Date: Thu, 24 Nov 2005 01:01:44 -0500 From: Kris Kennaway <kris@obsecurity.org> To: David Xu <davidxu@freebsd.org> Cc: current@FreeBSD.org, Kris Kennaway <kris@obsecurity.org> Subject: Re: 4BSD process starvation during I/O Message-ID: <20051124060143.GA15395@xor.obsecurity.org> In-Reply-To: <43850AB7.4000109@freebsd.org> References: <20051123201837.GA4822@xor.obsecurity.org> <438500BE.3020507@freebsd.org> <20051124000824.GA11032@xor.obsecurity.org> <43850AB7.4000109@freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
--9jxsPFA5p3P2qPhR Content-Type: text/plain; charset=us-ascii Content-Disposition: inline Content-Transfer-Encoding: quoted-printable On Thu, Nov 24, 2005 at 08:35:03AM +0800, David Xu wrote: > Kris Kennaway wrote: >=20 > >On Thu, Nov 24, 2005 at 07:52:30AM +0800, David Xu wrote: > >=20 > > > >>Kris Kennaway wrote: > >> > >> =20 > >> > >>>Perhaps this can be tweaked. > >>> > >>>Kris > >>> > >>>P.S. Please, no responses about how maybe someone could write a new > >>>scheduler that doesn't have this property. > >>> > >>> > >>> > >>> > >>> =20 > >>> > >>Can you try it again with FULL_PREEMPTION is turned on ? > >> =20 > >> > > > >Didn't really make a difference: > > > >=20 > > > This might only can be fixed when msleep no longer explicitly fiddles > thread priority, let scheduler fully control it. ULE doesn't seem to suffer from the "variance" problem nearly as badly on the simultaneous dd writing test I detailed earlier: ULE 268435456 bytes transferred in 6.738002 secs (39839028 bytes/sec) 268435456 bytes transferred in 6.755917 secs (39733385 bytes/sec) 268435456 bytes transferred in 6.828591 secs (39310518 bytes/sec) 268435456 bytes transferred in 7.173220 secs (37421891 bytes/sec) 268435456 bytes transferred in 7.359424 secs (36475064 bytes/sec) 268435456 bytes transferred in 7.452125 secs (36021330 bytes/sec) 268435456 bytes transferred in 7.657969 secs (35053087 bytes/sec) 268435456 bytes transferred in 7.848142 secs (34203695 bytes/sec) 268435456 bytes transferred in 8.061658 secs (33297798 bytes/sec) 268435456 bytes transferred in 8.159963 secs (32896652 bytes/sec) 268435456 bytes transferred in 8.176940 secs (32828351 bytes/sec) 268435456 bytes transferred in 8.184902 secs (32796416 bytes/sec) 4BSD 268435456 bytes transferred in 1.860367 secs (144291681 bytes/sec) 268435456 bytes transferred in 1.885677 secs (142354961 bytes/sec) 268435456 bytes transferred in 1.983243 secs (135351757 bytes/sec) 268435456 bytes transferred in 3.409261 secs (78737139 bytes/sec) 268435456 bytes transferred in 3.762066 secs (71353204 bytes/sec) 268435456 bytes transferred in 3.910402 secs (68646510 bytes/sec) 268435456 bytes transferred in 5.086984 secs (52769079 bytes/sec) 268435456 bytes transferred in 5.566311 secs (48225018 bytes/sec) 268435456 bytes transferred in 5.816347 secs (46151897 bytes/sec) 268435456 bytes transferred in 6.818567 secs (39368309 bytes/sec) 268435456 bytes transferred in 7.375191 secs (36397085 bytes/sec) 268435456 bytes transferred in 7.568077 secs (35469440 bytes/sec) Performance is lower on ULE though. Kris --9jxsPFA5p3P2qPhR Content-Type: application/pgp-signature Content-Disposition: inline -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.2 (FreeBSD) iD8DBQFDhVdHWry0BWjoQKURAm96AJ9J2uJCgnven7nCUyCWYVQIq7HljQCg2BVf GAVQcLK0aZvgD0185AiN5G8= =sYqT -----END PGP SIGNATURE----- --9jxsPFA5p3P2qPhR--
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20051124060143.GA15395>