1 |
On Sat, May 30, 2009 at 02:40:34PM +0100, Neil Bothwick wrote: |
2 |
> On Sat, 30 May 2009 12:06:04 +0200, Florian Philipp wrote: |
3 |
> |
4 |
> > Delaying commits with ext4 and/or laptop-mode will reduce the wear-down |
5 |
> > of your SSD but it might as well freeze your system when the actual |
6 |
> > commit takes place because these things tend to have a terribly low |
7 |
> > write performance. |
8 |
> |
9 |
> That may explain the pauses I get from time to time. Maybe shortening the |
10 |
> commit period will help. |
11 |
|
12 |
Couple of points regarding the pauses, SSDs, schedulers and ext3/ext4: |
13 |
|
14 |
* try ext4 with its delayed allocation. It should help with pauses |
15 |
* ext3 with data=writeback should help. Some security implications with |
16 |
data=writeback tho. So be careful if it is not a single user machine. |
17 |
* Deadline scheduler has more throughput than CFQ or anticipatory but it |
18 |
is totally unusable under load |
19 |
* A lot of patches to ext3 and ext4 for a/m pauses and SSDs. Some made |
20 |
it to kernel 2.6.30 I believe. |
21 |
* Try CFQ and NOOP as schedulers for SSDs for now. After the above |
22 |
patches, CFQ should be the better choice. |
23 |
|
24 |
Basically, a lot of changes to ext3/ext4 and schedulers at the moment. |
25 |
I would wait for at least kernel 2.6.31 before trying alternatives and |
26 |
making decisions. |
27 |
|
28 |
> Or I could try btrfs, which has an ssd mount option. |
29 |
|
30 |
Ugh. Even on-disk format is not finalized yet. |
31 |
|
32 |
-- |
33 |
Eray |