Re: FreeBSD 5.3b7and poor ata performance

Kevin Oberman wrote:
>>>OpenBSD 3.5:
>>>
>>>dd if=3D/mnt/aa of=3D/dev/null bs=3D1024k count=3D1024
>>>1024+0 records in
>>>1024+0 records out
>>>1073741824 bytes transferred in 18.743 secs (57287584 bytes/sec)
>>>
>>>
>>>FreeBSD 5.3rc1:
>>>
>>># dd if=3D/mnt/aa of=3D/dev/null bs=3D1024k count=3D1024
>>>1024+0 records in
>>>1024+0 records out
>>>1073741824 bytes transferred in 42.215597 secs (25434719 bytes/sec)
>>
>>BTW, others have asked that you not use dd as your "benchmark",
>>because it's not benchmarking anything relevant for real-world use of
>>the disk.
> 
> 
> While it is not typical of many "real-world" uses, I back up my laptop
> by copying the disk with dd(1). It used to take me about 40 minutes with
> V4. Now it takes just over an hour. This 25 minutes lost every day is a
> very real part of my world.

Same here. At work I use dd to backup last valid operating system
on the computer for easy recovery. It's a complement (not suplement)
of tape, library, etc backups. Also, OpenBSD "altroot" is a nice idea.


# If ROOTBACKUP is set to 1 in the environment, and
# if filesystem named /altroot is type ffs, on /dev/* and mounted "xx",
# use it as a backup root filesystem to be updated daily.
[ "X$ROOTBACKUP" = X1 ] && {
         rootdev=`awk '$2 == "/" && $1 ~ /^\/dev\// && $3 == "ffs" && \
             $4 ~ /rw/ \
                  { print substr($1, 6) }' < /etc/fstab`
         rootbak=`awk '$2 == "/altroot" && $1 ~ /^\/dev\// && $3 == "ffs" && \
             $4 ~ /xx/ \
                 { print substr($1, 6) }' < /etc/fstab`
         [ X$rootdev != X -a X$rootbak != X ] && {
                 sync
                 echo ""
                 echo "Backing up root filesystem:"
                 echo "copying /dev/r$rootdev to /dev/r$rootbak"
                 dd if=/dev/r$rootdev of=/dev/r$rootbak bs=16b seek=1 skip=1 \
                         conv=noerror
                 fsck -y /dev/r$rootbak
         }
}

> I also bothers me to see this significant deterioration on what appears
> to be a VERY simple disk activity. I'd like to benchmark buildworld, but
> doing that with different versions is not so easy. I might try building
> jdk. That is very "real-world".

It'd be interesting if you run some real world test. i.e. copy a big file
and compare times between F4 and F5