Manuel Bouyer --> netbsd-users (2007-06-27 21:45:54 +0200):
> On Wed, Jun 27, 2007 at 08:40:58PM +0200, Jukka Salmi wrote:
> > Jukka Salmi --> netbsd-users (2007-06-26 17:05:25 +0200):
> > [...]
> > > Hmm, any hints what could be the bottleneck here?
> > 
> > I've done some more rough testing to compare SATA vs. PATA read and
> > write speeds on wd, ccd and raid (levels 0 and 1), to and from the raw
> > devices and ffs respectively. I used dd(1) to write (read) 1 GiB to
> > (from) the device. The results are in MiB/s as reported by dd:
> > 
> > 
> >                         |-- SATA --|    |-- PATA --|
> >                         read   write    read   write
> > 
> >         wd:      raw      50   39         36   36
> >                  ffs      50   42         36   36
> > 
> >         ccd:     raw      91   40         72   72		
> >                  ffs      92   45         71   72				
> > 
> >         raid 0:  raw      90   40         71   73
> >                  ffs      90   45         68   70
> > 
> >         raid 1:  raw      49   21         35   36
> >                  ffs      49   23         35   36
> > 
> > 
> > Of course dd(1) is not a benchmark, but the results are - at least to
> > me - quite surprising: while read and write speed are always almost
> > identical on the PATA devices, reading from SATA devices is much faster,
> > but writing to them is much slower. Hmm. Not quite what I had expected.
> > I'd check with another sata controller if I had one...
> 
> Note that this can also be dependant on the drives. 
> The attached program will test the bandwidth between the drive's cache and the
> host. usage: ./tst /dev/rwd0d 10000 (increase last parameter if it completes
> too fast to give accurate results).

Thanks. The program prints about 85 MB/s for the SATA disks and about
113 MB/s for the PATA disks. Interesting...


Regards, Jukka

-- 
bashian roulette:
$ ((RANDOM%6)) || rm -rf ~