> dd -if /dev/sdE0/data -of /dev/sdF0/data -bs 1048576

i was thinking about your technique, and it occured to me
that this command is equvalent to
	for(;;)
		read(E0, buf, 1mb)
		write(F0, buf, 1mb)
but if you wrote it like this
	dd -if data -bs 64k -count 20000 |dd -bs 64k -of ../sda1/data
the read and write could be run in parallel, at the expense
of a buffer copy.  i didn't have anything except for some
very fast (120mb/s) sas disks to test with, but even they
showed 10% performance improvement. even at the expense
of copies
	0.01u 0.62s 11.63r 	 rc -c dd -if data -bs 64k -count 20000 -of ../sda1/data
	0.02u 0.97s 10.72r 	 rc -c dd -if data -bs 64k -count 20000|dd -bs 64k -of ../sda1/data
not all that impressive with my disks, perhaps this would
show more improvement on normal disks.

> fn chk {
>     for(i in sdE0 sdF0) dd -if /dev/$i/data -bs 1048576 -iseek $1
> -count 1 |md5sum
> }

i found this interesting, too.  i wrote a short program using the
threads library to do the reads and compares in parallel.  in
the process of writing that i realized that the md5sum is not
necessary.  a memcmp would do.  i finished the program up
(attached) and found that it performed pretty well.  giving me
~123mb/s.  that's about what these drives will do.  but i was
wondering why cmp would just work.  it occurred to me that
i could run the dds in parallel with a command like this
	cmp <{dd -if data -bs 64k -count 20000} <{dd -if ../sda1/data -bs 64k -count 20000}
surprisingly, this was just as fast on my setup as the specalized
program
	0.07u 0.04s 10.65r 	 8.out -n 20000 data ../sda1/data ...
	0.32u 0.26s 10.65r 	 cmp /fd/7 /fd/6

clearly if the compare is more involved, like sha1sum, it would
be more fruitful to use a modified version of the threaded program.
(unless you see a way of parallelizing the cmp part of that command
without byzantine contortions.)  i ran this test and found a surprising
speedup:
	0.06u 0.03s 13.65r 	 8.out -sn 20000 data ../sda1/data ...
i suspect there is something a bit amiss with time(1)'s accounting.

i suppose that a motivated person could write a book on parallel
programming with the shell.  tony hoar would be proud.

- erik