| 1 |
On 2009-01-24, Stroller <stroller@××××××××××××××××××.uk> wrote: |
| 2 |
|
| 3 |
> > 3. Does creating the swapfile on a journaled filesystem (e.g. |
| 4 |
> > ext3 or reiser) incur a significant performance hit? |
| 5 |
> |
| 6 |
> None at all. The kernel generates a map of swap offset -> disk |
| 7 |
> blocks at swapon time and from then on uses that map to perform |
| 8 |
> swap I/O directly against the underlying disk queue, bypassing all |
| 9 |
> caching, metadata and filesystem code. |
| 10 |
|
| 11 |
I supposed that the NT kernel does something similar. |
| 12 |
|
| 13 |
One implication of that is that the filesystem is then not |
| 14 |
allowed to move blocks around if they are part of an "active" |
| 15 |
swap file? Not that I'm aware of filesystems that shuffle |
| 16 |
blocks around while they're part of an open file, but one might |
| 17 |
imagine something like that happening as part of some sort of |
| 18 |
balancing algorithm. |
| 19 |
|
| 20 |
-- |
| 21 |
Grant |
Archives