Famous kernel developer Ingo Molnar said the following:

Atime updates are by far the biggest IO performance deficiency that Linux has today.  Getting rid of atime updates would give us more everyday Linux performance than all the pagecache speedups of the past 10 years, _combined_.

http://blogs.koolwal.net/2009/01/30/installing-linux-on-usb-part-4-noatime-and-relatime-mount-options/

And here's some of what the L man said about it:

http://lkml.org/lkml/2007/8/4/73

http://lkml.org/lkml/2007/8/10/200

Noatime is good, but remember that you will not be able to know when something was last accessed, so consider your needs carefully. Many people can do without atime, though you must be careful, as some things like backups, the popularity contest, and other things rely on it.

Be well informed!

I'm guessing this will also result in better battery life on a laptop as well.

It would be nice if some kind of atime cache was implemented so file accesses wouldn't cause disk IO.  ATime infomation could then be written at long intervals, and at unmount time.

This kerneltrap.org conversation has a similar suggestion.

http://kerneltrap.org/node/14148

This is in fact an old discussion which is already covered extensively.

One good reference concerning notaime (back in the year 2000!) was:

http://en.tldp.org/LDP/solrhe/Securing-Optimizing-Linux-RH-Edition-v1.3/chap6sec73.html

In the more recent days (2007), there was also a discussion available through Kerneltrap:

http://kerneltrap.org/node/14148

In any case, one could consider using the relatime flag by default to optimse disk IO performance not breaking application functionality.

Wiggum.

I know it sounds obvious, of course it does, not setting the atime of each file accessed or touched by a program sounds like a certain speedup for somethings.

 but the problem is, how much of an improvement is there?

 stats please! I'd like to see your reasoning about this further than "of course it would work" cause I am sure there are lots of things which "look" like they would improve performance but in real practice dont do any actual difference.

 Just look at STL containers like <vector> lots of people think that using them are slow, they say that "of course plain old arrays are faster!!!! they dont have all that STL  junk" yet actually, they dont realise that in reality, it's not faster at all and it's actually more bug prone and the same speed (if you have a good compiler).

 So! with that said: stats please! lets see your proof.

Hi there,

That sounds excellent, especially for people who dish out solutions on worn out hardware (for windows).

Simply putting a swap space on the another physical hard disk  does wonders especially if u have dual/multiple controllers (SCSI/SATA) etc..

I can safely say off the top of my head that performance gains are around 40% at times. (although this figure hasn´t been benchmarked thru standard benchmarkers)

I wonder if u have any experiences testing this in busy environment.

 -

Oz