[ltp] Re: HDD access avery 5 seconds?

[Chris Burkhardt]

Richard Neill wrote:
> 
> 
> Kiko Piris wrote:
>> On 14/08/2008 at 15:15 +0100, Richard Neill wrote:
>>
>>> Thanks - that's very useful. Just one thing puzzles me: why would
>>> ext3  flush the journal at all if the machine is idle; if no files
>>> have been  modified during that period, surely there's no data, nor
>>> updates to the  journal to be saved to disk?
>>
>> Mounting filesystems with noatime and nodiratime is crucial.
>>
>> If not, that means that every file read becomes also a write (filesystem
>> needs to update atimes).
>>
> 
> I've already go noatime everywhere.
> Do I need nodiratime too, or is nodiratime a subset of noatime?

nodiratime is implied by noatime. You do not need both. Interesting discussion
on the topic from a year ago:

http://kerneltrap.org/node/14148

> However, that isn't really the root cause. What I want to know is:
> How do I identify which *process* is accessing the filesystem (and
> thereby causing disk access) when the machine is idle.

powertop might help you track that down. (Although the iotop somebody else
mentioned is probably a better bet.)

I'd be interested to know what the culprit is when you discover it.

- Chris