> There are other, more significant changes between 1.6.2 and current,
> particularly in the way files are laid out on the disk. If you assume
> that writing is the bottleneck, your result suggests a huge win for
> the new dirpref code, but there isn't really enough information:
> we'd need to see OLD -> OLD and NEW -> NEW, and perhaps whether newly
> written files can be read back any faster from the old volume than the
> old files that were created under 1.6.2.
just to complete the information -  this computer started as 1.6.1 last 
autumn and almost immediatelly was turned into -current (for SMP support)

i must apologize for bad information, the situation is just reversed... 
i mixed 2 "labels" - from now i'll report it as SMALL (=OLD) and LARGE 
(=NEW)

160mb of data (values from my latest tests):
copy LARGE->LARGE	68.70s
move LARGE->LARGE	 0.01s

copy SMALL->SMALL	 8.25s
move SMALL->SMALL	 0.00s

copy SMALL->LARGE	56.10s
move SMALL->LARGE	63.37s

copy LARGE->SMALL	 8.42s
move LARGE->SMALL	 8.31s

 > Are the disks of the same model? Are you using softdeps? If
 > not, have you tried using softdeps?
both disks are exactly the same model. there is some difference in 
partitioning (LARGE disk is single partition, SMALL is divided in 3 
partitions).

if you need any information (dmesg, ...) or if i should do some other 
tests, just let me know.

> There's not enough information there to draw a conclusion. Do you mean
> that the command returns quickly, but the operation takes a long time
> to commit to disk? Are operations on the old volume really slower than
> they were under 1.6.2, or are they just slower than operations on new
> volumes? 
midnightcommander stuff: when i move file from new volume (LARGE) to the 
old one (SMALL), everything is smooth and fast - progress bar just goes 
on without stopping (reported average speed is something about 
16-18MB/second). when i do it the other way around, there are several 
stops during the moving of the file - i.e. progress bar freezes several 
times for several seconds and then continues (reported average speed is 
about 2-3MB/second). when the indicator hits 99-100% it stops for longer 
time.

i haven't tried softdeps, should i? i thought that i'd rather avoid it 
since i saw some problems with them were reported on the netbsd-current 
maillist, but if it is safe and give me some performance improvement, 
why not? :)
i guess i just need to rebuild my kernel with support for these?

regards,

-- 
Lubos
_@_"