gorn
Group: Members
Posts: 3
Joined: Sep. 2005 |
 |
Posted: Sep. 17 2005,09:50 |
 |
Groovy - looks like we nailed this one down. Thanks for all the good tips. Here's what I did:
1. Installed the real dd command as part of the gnu-utils.dsl. Downloaded the .dsl and simply copied to the root directory - DSL recognized and loaded at boot (a lot easier than what I remember having to do under Solaris).
Results using real dd:
time dd if=/dev/hda of=/dev/null bs=512 count=1953126
1953126+0 records in
1953126+0 records out
1000000512 bytes transferred in 327.886628 seconds (3049836 bytes/sec)
Original results using dd -> busybox:
time dd if=/dev/hda of=/dev/null bs=512 count=1953126
1953126+0 records in
1953126+0 records out
real 5m28.150s
user 0m0.690s
sys 4m4.080s
Although this showed almost no improvement in the test, I can't be sure of the real effect because I have never benchmarked the drive to drive copy (forgot to time the command and don't feel like waiting 7-8 hours to do so right now). I did have the impression the the drive was maxed out doing the sequential read using the real dd - with the busybox dd it seemed to sometimes cycle.
2. Enable the dma paramater on hda according to directions above
Results using real dd with dma activated:
time dd if=/dev/hda of=/dev/null bs=512 count=1953126
1953126+0 records in
1953126+0 records out
1000000512 bytes transferred in 28.466377 seconds (35129181 bytes/sec)
Performance improvement was at least a factor of 10!
I'll do a real life test with dma activated for both hda and hdc, then post results.
Thanks again to all for the help.
|
.
DSL Market ,
Great VPS hosting provided by Tektonic
