> You might try using even larger blocks, like 32k, or playing with the
> parameters to "tunefs(8)". 
i tried even 64k blocks, but it didn't help much, in fact the situation 
was slightly worse than for 32k blocks...

> It might not be a bad idea to go ahead and
> build a file system on the new disk using a 1.6.2 floppy, just to rule
> out the hardware completely. 
i did - with 16k blocks the 'cpg' value was something like 368, but the 
disk was slow anyway... the same with 8k blocks and cpg 86 (that's the 
same value as for the faster drive) - so should i blame hardware? but 
why would be raw writing to /dev/rwd1 fast if the drive wouldn't be ok?

i tried with dkctl again and it reports write cache enabled. however, i 
think there's something wrong here. i can see that reading is fast, but 
for writing it seems to fill the buffer and waiting 3seconds (i guess 
the buffer has size of approx. 16mb, since the "chokes" are present each 
8% of the testing 200mb large file)

is there any program to properly test what's going on?

thanks,

-- 
Lubos
_@_"