The numbers for the test in case are currently about 34% cpu for the
disk to /dev/null and about 64% for disk to disk.
I can't remeasure the tcp numbers I don't have the hardware anymore.
The same test on a rather busy freebsd machine yields 2.6% cpu for the
disk to disk case.
The big question is how do you measure the CPU percentage. If you do so
using the time command, you will get bad numbers because of differences
in how different OS's charges the CPU time spent in interrupt handlers
--- to which process should they get charged? If under BSD all of the
CPU time spent in the SCSI interrupt handlers were charged to the idle
process, that would handily explain why FreeBSD yielded 2.6% cpu for the
disk-to-disk case.
The only valid way to measure CPU utilization is to start a process
which times how many interations of a busy loop can be executed per unit
time while the system is quiscient, and then compare that with how many
interations of the busy loop is executed while the disk copy (or serial
transfer, or whatever) is taking place.
Since I can't currently do the tests, I will have to retract my
numbers in the face of people who have current test results that show
different numbers.
Were you original tests done in the fashion I described above, or did
you just use the time command? If you just used the time command, the
results were never accurate to begin with, whether or not the results
were using current versions of the two OS's.
- Ted
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/