Re: FreeBSD 5.3b7and poor ata performance

At 12:02 PM 19/10/2004, fandino wrote:

># dd if=/dev/ad4 of=/dev/null bs=1024k count=1024
>1024+0 records in
>1024+0 records out
>1073741824 bytes transferred in 30.606110 secs (35082597 bytes/sec)


Its strange that you do not see any difference in speed when reading from 
the device directly vs a file ? I see quite a difference in my setup on a 
RELENG_5 box and not so much a difference on a RELENG_4 box.  Perhaps as 
someone suggested, the GEOM layer ?

e.g. create a file with junk larger than your RAM. (the machine below has 256M)

On FreeBSD 915.sentex.ca 5.3-STABLE FreeBSD 5.3-STABLE #1: Mon Oct 18

915# dd if=/dev/urandom of=/usr/tmp/b bs=1024k count=1024
1024+0 records in
1024+0 records out
1073741824 bytes transferred in 40.103750 secs (26774100 bytes/sec)

Now, for the read speed

915# dd if=/dev/ad2s1d of=/dev/null bs=1024k count=1000
256+0 records in
256+0 records out
268435456 bytes transferred in 4.606747 secs (58270065 bytes/sec)
915# dd if=/dev/ad2 of=/dev/null bs=1024k count=1000
1000+0 records in
1000+0 records out
1048576000 bytes transferred in 18.049106 secs (58095730 bytes/sec)
915# dd if=/usr/tmp/b of=/dev/null bs=1024k count=1000
1000+0 records in
1000+0 records out
1048576000 bytes transferred in 29.876008 secs (35097594 bytes/sec)
915# dd if=/usr/tmp/b of=/dev/null bs=1024k count=1000
1000+0 records in
1000+0 records out
1048576000 bytes transferred in 29.952459 secs (35008011 bytes/sec)
915#

58 vs 35.

Yet on a RELENG_4 machine, the difference is not nearly as 
acute.  Considering the RELENG_4 box is a Celeron 1000 blasting a ad0: 
38166MB <ST340014A> [77545/16/63] at ata0-master UDMA100 I am surprised by 
the results.

proxy-a# dd if=/dev/ad0s1e of=/dev/null bs=1024k count=1000
1000+0 records in
1000+0 records out
1048576000 bytes transferred in 20.884694 secs (50207870 bytes/sec)
proxy-a#
proxy-a# dd if=/usr/tmp/b of=/dev/null bs=1024k count=1000
1000+0 records in
1000+0 records out
1048576000 bytes transferred in 24.908032 secs (42097907 bytes/sec)


         ---Mike