| 1 |
Bryan, |
| 2 |
|
| 3 |
Here is John's reply. |
| 4 |
|
| 5 |
Bryan Green wrote: [Thu Aug 31 2006, 09:02:40PM EDT] |
| 6 |
> I'm looking at the kernel config right now. |
| 7 |
> Is it dangerous to change some of the kernel config options? |
| 8 |
> Are there specific options that the lustre patches depend on? |
| 9 |
> For example, since I'll be running this on a box with interactive graphics, I'd like to |
| 10 |
> change the Preemption Model to "Preemptible Kernel". |
| 11 |
|
| 12 |
In general, my experience is that you're pretty safe. There are exceptions. |
| 13 |
Obviously you have to have modules enabled etc. You also have to leave |
| 14 |
enabled the SD_IOSTATS one, which I believe shows up as the legacy /proc/sd |
| 15 |
filesystem, somewhat non-intuitively. |
| 16 |
|
| 17 |
The short answer for that specific one is that CFS claims that you're not |
| 18 |
supposed to change the preemption model, but I've tried it with various values |
| 19 |
(some by accident) and never observed a problem. In my travels through the |
| 20 |
code, I've observed that they *do* have all kinds of locking happening all |
| 21 |
over the place, and changing that kernel setting *does* cause fairly radical |
| 22 |
changes in the locking stuff, so I would be willing to believe that lustre |
| 23 |
does something bad if you change it. But I've never observed a problem. I'd |
| 24 |
say try it, and if it misbehaves, turn it back :-} |
| 25 |
|
| 26 |
-- |
| 27 |
Joel Martin (kanaka) |
Archives