Hi, On Tue, 18 Dec 2007, Ingo Juergensmann wrote:
> On Mon, Dec 17, 2007 at 11:27:42PM +0100, Roman Zippel wrote: > > > > I've upgraded the following machines lately to kernel 2.6.23-m68k and > > > etch-m68k: > > You might want to test 2.6.22 as well. The new scheduler hasn't exactly > > been an improvement for us, here are some lmbench numbers: > > Context switching - times in microseconds - smaller is better > > ------------------------------------------------------------------------- > > Host OS 2p/0K 2p/16K 2p/64K 8p/16K 8p/64K 16p/16K 16p/64K > > ctxsw ctxsw ctxsw ctxsw ctxsw ctxsw ctxsw > > --------- ------------- ------ ------ ------ ------ ------ ------- ------- > > sodom Linux 2.6.22- 52.4 125.9 123.5 158.8 153.6 174.8 153.6 > > sodom Linux 2.6.22- 49.9 142.4 126.5 158.5 156.8 176.6 152.9 > > sodom Linux 2.6.23- 112.0 176.6 159.8 256.9 216.4 268.8 242.1 > > sodom Linux 2.6.23- 112.4 177.5 164.8 259.5 232.5 263.9 240.9 > > sodom Linux 2.6.23- 133.3 206.6 189.8 280.1 269.8 297.9 267.6 > > Oooops... that's not really an improvement... anyway, I've chosen 2.6.23 as > it was the latest and the easiest to checkout from the m68k cvs for me. You can use "-r m68k-2_6_22" with update/checkout. > And it contained the SCSI patches for WarpEngine and A4000T SCSI.... This should be in 2.6.22 as well and I wouldn't mind creating a branch to backport smaller driver fixes, so we can use that for a while until the scheduler issues have been sorted out. > I assume that the above values are valid for the default scheduler, but how > about other schedulers? > > CONFIG_IOSCHED_NOOP=y > CONFIG_IOSCHED_AS=y > CONFIG_IOSCHED_DEADLINE=y > CONFIG_IOSCHED_CFQ=y > CONFIG_DEFAULT_IOSCHED="cfq" That's the io scheduler not the task scheduler. bye, Roman -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
