From `man atop`:
Code:
DSK Disk utilization.
Per active disk one line is produced, sorted on disk activity.
Such line shows the name of the disk (e.g. hda or sda), the busy
percentage i.e. the portion of time that the disk was busy han-
dling requests (âbusyâ), the number of read requests issued
(âreadâ), the number of write requests issued (âwriteâ) and the
average number of milliseconds needed by a request (âavioâ) for
seek, latency and data transfer.
That's a sufficient answer for me. If you want to know more about how it calculates that (that seems beyond the abilities of /proc), you may need to dig into the source or dev mailing lists.