Poor SCSI disk preformance

[Derek Marcotte]

> I wouldn't say that dd is the greatest benchmarking tool. You
may want to
> try benchmarks/rawio.
I'll check that out just for kicks, but I _actually want_ to
write zeros to the drive first, not just as a benchmark.  The
reasoning for this is that I'm trying to create a dedicated box
to format HDDs in parallel.  I wish to first zero the drives to
make data recovery without an electron microscope difficult.

Then, to test for bad sectors I do a checksum of the number of
zero bytes written to disk, and then I read back from the disk
and compare checksums. Not exactly an extensive test, and perhaps
there is a verify option or trick in dd that I'm not aware of.  I
think that this would catch any blatantly bad drives...

If not, there are 2 full disk operations that should be going
faster.

Actually, just for kicks:

# dd if=/dev/da0 of=/dev/null bs=128k &
[1] 1839
# iostat -K -w 1 da0
      tty             da0             cpu
 tin tout  KB/t tps  MB/s  us ni sy in id
   1   42 64.00 607 37.92   1  0  1  0 98
   0   43 64.00 222 13.87   0  0  2  0 98
   0   43 64.00 223 13.92   0  0  0  2 98
   0   42 64.00 224 13.98   0  0  2  0 98
   0   43 64.00 222 13.86   0  0  3  0 97
   0   43 64.00 223 13.92   0  0  1  2 98
   0   43 64.00 223 13.92   0  0  2  1 97
   0   42 64.00 223 13.92   0  0  3  0 97
   0   43 64.00 223 13.92   0  0  1  0 99

Seems to give me the performance that I expect...


> Also, try monitoring diffferent types of transfers to
> and from another physical disk with iostat.
Actually, interestingly enough, when I copy a file, or do a
newfs_msdos I only get 0.06-0.89MB/s transfers, which is what
first tipped me off to the problems...  Obviously less
acceptable...