| 1 |
On 24 Jan 2009, at 17:22, Grant Edwards wrote: |
| 2 |
> I still can't believe that Windows does it's swapping using a |
| 3 |
> normal filesystem -- and by default it's the same filesystem |
| 4 |
> used for system and application files. It seems like the |
| 5 |
> filesystem code would end up being a serious bottleneck. |
| 6 |
|
| 7 |
> 3. Does creating the swapfile on a journaled filesystem (e.g. |
| 8 |
> ext3 or reiser) incur a significant performance hit? |
| 9 |
|
| 10 |
None at all. The kernel generates a map of swap offset -> disk |
| 11 |
blocks at swapon time and from then on uses that map to perform |
| 12 |
swap I/O directly against the underlying disk queue, bypassing all |
| 13 |
caching, metadata and filesystem code. |
| 14 |
|
| 15 |
http://lkml.org/lkml/2005/7/7/326 |
| 16 |
|
| 17 |
Stroller. |
Archives