Description of problem:

the readahead program currently doesn't use O_NOATIME while opening the files to
read ahead... this has 2 effects
1) the atime changes, potentially impacting backup and other tools that think
files are used more than they maybe are
2) about 5 seconds afterwards, kjournald starts writing the new dirty inodes out
to disk, slowing things down unneeded.

Fix is easy, just add "| O_NOATIME" to the open flags...