/proc/partitions statistics question

[Jean-Luc Fontaine <jfontain@free.fr>]

I have been trying to make sense of the units used in the 
/proc/partitions statistics.
I thought read and write I/O were in kilobytes, but not according to the 
following test on my 2.4.3 kernel machine:

# /proc/partitions columns:
# major minor #blocks name rio rmerge rsect ruse wio wmerge wsect wuse 
running use aveq

$ fgrep hdb1 /proc/partitions
$ dd if=/dev/zero of=ttt bs=1M count=1024 2>/dev/null
$ sync; sync; sync
$ fgrep hdb1 /proc/partitions
    3    65   40209088 hdb1 7107 18180 202278 384530 7925 19005 215616 
215330 0 46240 599860
    3    65   40209088 hdb1 7134 18218 202798 386510 24566 265384 
2319776 2977060 0 102000 3363570
$ ls -lk ttt
-rw-r--r--    1 root     root      1048576 Dec 13 15:56 ttt
$ fgrep hdb1 /proc/partitions
$ dd if=ttt of=/dev/null 2>/dev/null
$ sync; sync; sync
$ fgrep hdb1 /proc/partitions
    3    65   40209088 hdb1 7152 18222 202974 386650 24581 265397 
2320000 2977060 0 102140 3363710
    3    65   40209088 hdb1 23567 263977 2300334 555980 24668 265441 
2321048 2978280 0 171380 3534260

(24566 - 7925) = 16641 were written, whereas
(23567 - 7152) = 16415 were read
I would have expected something near 1048576 (kilobytes),
but the results seem to be roughly 64 times less...

I would really appreciate an explanation, as I use those statistics
in a monitoring program and for filesystems performance tests.

Many thanks for your help!

-- 
Jean-Luc Fontaine  mailto:jfontain@free.fr  http://jfontain.free.fr/