| 1 |
Am 13.08.2012 19:14, schrieb Florian Philipp: |
| 2 |
> Am 13.08.2012 16:52, schrieb Michael Mol: |
| 3 |
>> On Mon, Aug 13, 2012 at 10:42 AM, Michael Hampicke |
| 4 |
>> <mgehampicke@×××××.com <mailto:mgehampicke@×××××.com>> wrote: |
| 5 |
>> |
| 6 |
>> Have you indexed your ext4 partition? |
| 7 |
>> |
| 8 |
>> # tune2fs -O dir_index /dev/your_partition |
| 9 |
>> # e2fsck -D /dev/your_partition |
| 10 |
>> |
| 11 |
>> Hi, the dir_index is active. I guess that's why delete operations |
| 12 |
>> take as long as they take (index has to be updated every time) |
| 13 |
>> |
| 14 |
>> |
| 15 |
>> 1) Scan for files to remove |
| 16 |
>> 2) disable index |
| 17 |
>> 3) Remove files |
| 18 |
>> 4) enable index |
| 19 |
>> |
| 20 |
>> ? |
| 21 |
>> |
| 22 |
>> -- |
| 23 |
>> :wq |
| 24 |
> |
| 25 |
> Other things to think about: |
| 26 |
> |
| 27 |
> 1. Play around with data=journal/writeback/ordered. IIRC, data=journal |
| 28 |
> actually used to improve performance depending on the workload as it |
| 29 |
> delays random IO in favor of sequential IO (when updating the journal). |
| 30 |
> |
| 31 |
> 2. Increase the journal size. |
| 32 |
> |
| 33 |
> 3. Take a look at `man 1 chattr`. Especially the 'T' attribute. Of |
| 34 |
> course this only helps after re-allocating everything. |
| 35 |
> |
| 36 |
> 4. Try parallelizing. Ext4 requires relatively few locks nowadays (since |
| 37 |
> 2.6.39 IIRC). For example: |
| 38 |
> find $TOP_DIR -mindepth 1 -maxdepth 1 -print0 | \ |
| 39 |
> xargs -0 -n 1 -r -P 4 -I '{}' find '{}' -type f |
| 40 |
> |
| 41 |
> 5. Use a separate device for the journal. |
| 42 |
> |
| 43 |
> 6. Temporarily deactivate the journal with tune2fs similar to MM's idea. |
| 44 |
> |
| 45 |
> Regards, |
| 46 |
> Florian Philipp |
| 47 |
> |
| 48 |
|
| 49 |
Trying out different journals-/options was already on my list, but the |
| 50 |
manpage on chattr regarding the T attribute is an interesting read. |
| 51 |
Definitely worth trying. |
| 52 |
|
| 53 |
Parallelizing multiple finds was something I already did, but the only |
| 54 |
thing that increased was the IO wait :) But now having read all the |
| 55 |
suggestions in this thread, I might try it again. |
| 56 |
|
| 57 |
Separate device for the journal is a good idea, but not possible atm |
| 58 |
(machine is abroad in a data center) |
Archives