| 1 |
On Sun, Nov 28, 2010 at 7:39 PM, walt <w41ter@×××××.com> wrote: |
| 2 |
|
| 3 |
> I'm confused about which of all these various mechanisms apply to single-cpu |
| 4 |
> machines. AFAICT Con's BFS (e.g.) is really a CPU scheduler and doesn't |
| 5 |
> affect |
| 6 |
> single-cpu machines very much. What about CFQ and group scheduling? |
| 7 |
> Others? |
| 8 |
> |
| 9 |
> Thanks for any clues. |
| 10 |
|
| 11 |
Don't mix them, |
| 12 |
|
| 13 |
CFS --> upstream official CPU scheduler (also supports cgroups, that |
| 14 |
got used in the 200line patch, which is useless imo) |
| 15 |
CFQ --> upstream official I/O (disk) scheduler (afaik the only one |
| 16 |
that supports "ionice") |
| 17 |
|
| 18 |
|
| 19 |
BFS --> ck's CPU scheduler |
| 20 |
(I don't know what i/o scheduler ck's patchset uses) |
| 21 |
|
| 22 |
|
| 23 |
Anyway, the problem with the long pauses under disk usage is not |
| 24 |
related to any scheduler at all, and it is a "page cache management" |
| 25 |
problem that the linux kernel has (in all new versions). |
| 26 |
|
| 27 |
CK's patchset includes optimizations to page cache management. (Not |
| 28 |
related to any scheduler) so it is worth a try. That's it. |
Archives