I've been looking into some of the reasons that a nearly idle Linux system might perform a surprising number of disk writes.

The latest one I've come across is that writing to (for instance) /dev/ttyS0 causes the mtime to change, so the inode gets dirty and is flushed to disk on the next kernel buffer timeout.

But I'm only seeing this disk write under a 2.2.x kernel, not a 2.4.x kernel. That is, the mtime changes in both cases, but the change doesn't result in a disk write under the 2.4.x kernel.

Is there a difference in device inode handling between 2.2.x and 2.4.x that would account for this?

Thanks.