Re: [patch] CFS scheduler, -v18
On Tue, Jul 03, 2007 at 10:10:05AM +0200, Keith Packard wrote: > On Tue, 2007-07-03 at 09:22 +0200, Ingo Molnar wrote: > > > which allows xterm-spam (attached) to easily flood the xterm (without > > any scrolling that would act as a throttle) and the xterm to flood Xorg. ...deleting the offhand remarks ...here's a little content: > And, as the kernel will not buffer more than 4095 bytes from a pty, > there isn't any way it will ever read 4096 bytes. The sched_yield() call was added to address problems in an earlier 2.6.x kernel (and the report then stated that it could read more than 4096 because it would have to perform several reads to keep up with a burst of output). ...deleting the other offhand remarks -- Thomas E. Dickey http://invisible-island.net ftp://invisible-island.net pgpk3381nfz3p.pgp Description: PGP signature
Re: [patch] CFS scheduler, -v18
On Tue, Jul 03, 2007 at 10:40:07AM +0200, Ingo Molnar wrote: > yeah, i use gnome-terminal exclusively. But testers looking for CFS > regressions do run every shell on the planet :-) ...and people running older kernels get different results (no surprise) fwiw, I ran 'top' on 5 terminals with xterm-spam running concurrently on 2.6.15 (rxvt, pterm, xterm, konsole and gnome-terminal). For that case, gnome-terminal was definitely the slowest, and used the most CPU time (more than a factor of three slower than xterm). konsole was about 2.5, pterm was about the same as xterm, and rxvt about half the CPU (ymmv). > > gnome-terminal is also faster all around (at least on my box): -- Thomas E. Dickey http://invisible-island.net ftp://invisible-island.net pgp6BczIooy34.pgp Description: PGP signature
Re: [patch] CFS scheduler, -v18
On Tue, Jul 03, 2007 at 10:40:07AM +0200, Ingo Molnar wrote: yeah, i use gnome-terminal exclusively. But testers looking for CFS regressions do run every shell on the planet :-) ...and people running older kernels get different results (no surprise) fwiw, I ran 'top' on 5 terminals with xterm-spam running concurrently on 2.6.15 (rxvt, pterm, xterm, konsole and gnome-terminal). For that case, gnome-terminal was definitely the slowest, and used the most CPU time (more than a factor of three slower than xterm). konsole was about 2.5, pterm was about the same as xterm, and rxvt about half the CPU (ymmv). gnome-terminal is also faster all around (at least on my box): -- Thomas E. Dickey http://invisible-island.net ftp://invisible-island.net pgp6BczIooy34.pgp Description: PGP signature
Re: [patch] CFS scheduler, -v18
On Tue, Jul 03, 2007 at 10:10:05AM +0200, Keith Packard wrote: On Tue, 2007-07-03 at 09:22 +0200, Ingo Molnar wrote: which allows xterm-spam (attached) to easily flood the xterm (without any scrolling that would act as a throttle) and the xterm to flood Xorg. ...deleting the offhand remarks ...here's a little content: And, as the kernel will not buffer more than 4095 bytes from a pty, there isn't any way it will ever read 4096 bytes. The sched_yield() call was added to address problems in an earlier 2.6.x kernel (and the report then stated that it could read more than 4096 because it would have to perform several reads to keep up with a burst of output). ...deleting the other offhand remarks -- Thomas E. Dickey http://invisible-island.net ftp://invisible-island.net pgpk3381nfz3p.pgp Description: PGP signature
