Rogier Wolff <R.E.Wolff@BitWizard.nl> writes:

> The way to fix this would be to be able to assign a higher cache
> priority (*) to the blocks in the FAT, and to read more than just 4k
> per seek to the FAT.

I tried it by attached *stupid* patch.

copying 500M data

	root@devron (x)[1014]# time dd if=/dev/hda6 bs=1M count=500 > /dev/null
	500+0 records in
	500+0 records out
	524288000 bytes transferred in 44.988916 seconds (11653715 bytes/sec)
	
	real    0m45.011s
	user    0m0.008s
	sys     0m8.723s

2.5.70-bk9
	root@devron (a)[1032]# time dd if=file bs=1M count=500 > /dev/null
	500+0 records in
	500+0 records out
	524288000 bytes transferred in 75.510951 seconds (6943205 bytes/sec)
	
	real    1m15.538s
	user    0m0.015s
	sys     0m16.493s

2.5.70-bk9+patch
	root@devron (a)[1024]# time dd if=file bs=1M count=500 > /dev/null
	500+0 records in
	500+0 records out
	524288000 bytes transferred in 52.034399 seconds (10075796 bytes/sec)
	
	real    0m52.041s
	user    0m0.006s
	sys     0m17.594s

You're right. It seems that that optimization has sufficient effect to
non fragment file. Thanks.
-- 
OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>