1 |
On Sun, Jul 31, 2011 at 1:39 PM, Peter Humphrey |
2 |
<peter@××××××××××××××.org> wrote: |
3 |
> On Sunday 31 July 2011 17:05:39 Michael Mol wrote: |
4 |
> |
5 |
>> However, if it's doing that, then it probably has something it needs to |
6 |
>> write to disk. That might be metadata updates. |
7 |
> |
8 |
> What, at least once a minute? While the system's idling, waiting for |
9 |
> something to do? Doesn't sound likely to me. |
10 |
> |
11 |
>> Have you tried adding things to your mount parameters like 'noatime' or |
12 |
>> 'relatime'? |
13 |
> |
14 |
> I've been specifying 'noatime' on all partitions for several years now; it's |
15 |
> automatic behaviour on my part. |
16 |
> |
17 |
>> What about "data=writeback"? |
18 |
> |
19 |
> I don't like the sound of the warning in the man page. |
20 |
> |
21 |
> Thanks for the ideas. So far I'm inclining to the reformatting I mentioned. |
22 |
|
23 |
Here's what I think is happening: |
24 |
|
25 |
ext3/ext4 is not going to arbitrarily poll writes to disk without |
26 |
there being something to write. Some program, somewhere on your system |
27 |
is doing something that involves modifying a file. Any filesystem that |
28 |
provides guarantees about disk integrity is going to get that data to |
29 |
a physically persistent state ASAP. That's why we have journaled |
30 |
filesystems in the first place: to speed that up. |
31 |
|
32 |
So, with the same application and configuration set, you're going to |
33 |
see the same behavior on any filesystem which provides such |
34 |
guarantees. |
35 |
|
36 |
You're perfectly welcome to reformat if you're so inclined; it really |
37 |
sounds like you're simply more comfortable (or more interested in) |
38 |
reiserfs. If you perceive that that solve your problem, great--but I |
39 |
don't think that would really solve the underlying technical issue. |
40 |
|
41 |
What you really want to do is find some way to log what's actually |
42 |
driving the data writes. If it were a particular app, it'd be as |
43 |
simple as launching the app via strace and analyzing the output. I |
44 |
don't know how one would do that system-wide, though. Perhaps someone |
45 |
else might have ideas. |
46 |
|
47 |
-- |
48 |
:wq |