On 8/6/05, Paul Kölle <[EMAIL PROTECTED]> wrote: > BTW: I did some research on IO scheduling when emerging stuff caused > squid to starve and made a lot of our users very unhappy. I was very > surprised that there was no per process IO scheduling already in the > kernel. There have been patches around in 2.6.11-mm_pre? prereleases > along with a userspace tool to set the IO nice level. However, I don't > know if that work will go on or if the results will show up in mainline > ever.
CFQ has been able to do this for a while with the supplemental "timeslice" support patches. The implementation is rather elegant, I/O priorities may be assigned proportionally in accordance with the nice level of the process issuing the requests. I've tested it on several occasions on my desktop and it seems to work very effectively indeed. So, in conjunction with these patches, the problem you mentioned might have been solved simply by renicing emerge. The -ck kernel has included this support for a while and hopefully this work will go mainline in the near future. I know that Andrew Morton mentioned "CFQ3" as a possible candidate for inclusion in 2.6.13. Here's the original check-in message from the CFQ author (Jens Axboe): "This patch updates the CFQ io scheduler to the new time sliced design (cfq v3). It provides full process fairness, while giving excellent aggregate system throughput even for many competing processes. It supports io priorities, either inherited from the cpu nice value or set directly with the ioprio_get/set syscalls. The latter closely mimic set/getpriority." Very interesting stuff, I'm sure you'll agree. If you fancy playing with it then just pull the following patches from http://ck.kolivas.org/patches/2.6/2.6.12/2.6.12-ck5/patches/ : * cfq-2.6.12-mm1.patch * cfq-ts-2.diff * cfq-ts-4.diff The 2nd and 3rd patches just aggregate some of the additional fixes which are also in the -mm tree. The -ck list is usually a good place to discuss this because almost all users of this patchset use the CFQ elevator and Jens keeps an eye on the list accordingly. Cheers, --Kerin Millar -- [email protected] mailing list
