2009/10/20 James Cameron <quozl@us.netrek.org>
On Tue, Oct 20, 2009 at 09:39:50AM +0800, Ray Rashif wrote:I'm not sure that is right; the way you put it.
> Since on a Linux system files are written to almost every minute,
> recording atime everytime an access occurs increases the disk I/O.
Perhaps you mean "Since on a Linux system files are read from almost
every minute, recording atime every time a read occurs increases the
write disk I/O."
atime on a file is changed whenever the file is read. The change is
buffered in memory for a while; usually no more than a minute, then is
written out to disk.
So atime causes at least one disk write I/O for a set of disk read I/Os.
It certainly does impact performance ... if it is not needed.
Another method is to place these files in a separate filesystem with
> I need to know access times for some files so I don't quite like that.
> I use relatime: http://kerneltrap.org/node/14148
atime enabled.