> ... hdparm uses disk raw
> I/O calls to measure performance and the 2.3.X kernels have had the disk
> code re-worked to improve overall system response while heavy disk
> I/O is occurring. You need to create your own test say by using the "time"
> command and "dd" to read a file of a known size then, calculte the
> transfer rate from the time report.
it's certainly true that the block layer has been reworked
(it now attempts to prevent starvation.) however, it's NOT
true that hdparm does anything different than "dd if=/dev/hda"
kind of hacks. it doesn't use any special calls, and it screws up
its own measurement by touching every sector that it reads
(thus making reported performance more CPU/cache/bus/dram
dependent, and penalizing DMA/UDMA transfers...)
in other words, use hdparm for nothing more than a sanity check;
it's not a decent benchmark. for sane benchmarking, try bonnie/iozone
on a file at least 3x the size of ram (and ignore the %CPU and per-char
columns in bonnie). on my systems, I'm seeing performance that's
approximately the same as in the 2.2 era.
regards, mark hahn.
-
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/
This archive was generated by hypermail 2b29 : Fri Apr 07 2000 - 21:00:10 EST