| 1 |
Willie Wong writes: |
| 2 |
|
| 3 |
> When the filesystem fills up, services can start failing left and |
| 4 |
> right because they cannot write logs, cannot write temp files, etc. At |
| 5 |
> this point human intervention is necessary: root has to log in and |
| 6 |
> clear out the disk. But if the $ROOT filesystem is completely full, |
| 7 |
> one may not even be able to log in and/or that one cannot do any sort |
| 8 |
> of maintenance that is needed. So you have some sort of circularity. |
| 9 |
> (In which case you have to reboot, perhaps using another medium...) |
| 10 |
> |
| 11 |
> The way out is to reserve some breathing room for root so that when |
| 12 |
> everybody else is having problems he can still get in and fix the |
| 13 |
> problem. |
| 14 |
> |
| 15 |
> The 5% is historical from days when disks are much smaller. If you |
| 16 |
> have a sensible partition scheme you only really need to reserve the |
| 17 |
> blocks on the $ROOT filesystem. If the partition in question (IIRC) is |
| 18 |
> only for /home, then you can just turn off the reserved blocks all |
| 19 |
> together. |
| 20 |
|
| 21 |
Isn't another purpose of those 5% the reduction of fragmentation that |
| 22 |
occurs more when there is few free space left? Although I also reduce ift |
| 23 |
on very large partitions. But I never set it to exactly zero. |
| 24 |
|
| 25 |
Wonko |
Archives