David,
thank you for the explanation, specially the "FireWire is much faster"
bit ;). Yes, indeed it is much faster then USB 1.x.
to clear up some variables:
The machine in question has 8 internal SCSI drives, the 4 GB file
resides on one of the drives. It doesn't have a DVD drive that I could
use for the test, and I suppose the DVD drive would be slower than the
external drive anyway.
The external drive only uses one of the two possible connectors at any
one time and I completely powered both, the server and the drive, down,
before conducting the test.
The external drive was blank when I started this test (mke2fs ...).
The pm_10.d2 file is a single large (4 GB) database extent of a database
that is completely shut down. You might say that it is equivalent to a 4
GB DVD ISO image, except for the content, which, for this test, should
not be relevant.
I have since tested this on another machine, an IBM eServer xSeries 342
running CentOS 4.4, with similar results. I don't think this issue is
only related to FC, I think it's related to Linux in general. I don't
have Windoze installed on a server, but I will test this on my other
desktop box which has built-in as well as another FireWire card. But
that will only happen tomorrow.
Best regards,
Chris
David Timms wrote:
Chris Ruprecht wrote:
Hello,
I have a Dell PowerEdge 2500 server with FC6 + latest updates installed.
For backup, I want to use 2 external disk drives, 300 GB each (all
internal stuff is SCSI). I bought a USB 2.0/FireWire combo card and
am running a few tests and to my great surprise, USB 2 (480 mbps) is
way faster than FireWire 400 (400 mbps). I get this from hdparm:
Connected as USB 2.0 drive:
hdparm -t -T /dev/sdd
/dev/sdd:
Timing cached reads: 948 MB in 2.00 seconds = 473.56 MB/sec
Timing buffered disk reads: 76 MB in 3.03 seconds = 25.11 MB/sec
Connected as FireWire 400 drive:
hdparm -t -T /dev/sdd
/dev/sdd:
Timing cached reads: 948 MB in 2.00 seconds = 473.05 MB/sec
Timing buffered disk reads: 32 MB in 3.07 seconds = 10.44 MB/sec
Notice this is only looking at read speeds.
To make sure that hdparm is not messing with me, as the only
reference I find in its man page is relating to IDE and SCSI drives,
I tested some simple copy command of a 4 GB file. I did each command
twice in a row, to avoind any skewing of the results by anything in
OS buffers.
USB 2.0 Mode:
time cp -v /home/db/pm/pm_10.d2 /media/extd/xxx3
`/home/db/pm/pm_10.d2' -> `/media/extd/xxx3'
real 3m20.819s
user 0m0.988s
sys 0m54.483s
[root@mars ~]# time cp -v /home/db/pm/pm_10.d2 /media/extd/xxx4
`/home/db/pm/pm_10.d2' -> `/media/extd/xxx4'
real 3m18.799s
user 0m1.048s
sys 0m54.791s
FireWire Mode:
time cp -v /home/db/pm/pm_10.d2 /media/extd/xxx1
`/home/db/pm/pm_10.d2' -> `/media/extd/xxx1'
real 7m33.109s
user 0m1.196s
sys 0m52.439s
[root@mars ~]# time cp -v /home/db/pm/pm_10.d2 /media/extd/xxx2
`/home/db/pm/pm_10.d2' -> `/media/extd/xxx2'
real 9m13.851s
user 0m1.108s
sys 0m55.191s
You are causing the data to be duplicated on the disk; this might slow
down with each subsequent write due to disks generally being fastest
transfer at the begining, and speed trailing of by 20% toward the end
of the disk.
I used the same drive, a Maxtor 320 GB drive that was on sale
recently, which is mounted inside an Adaptec external USB/FireWire
combo case.
Any ideas why FireWire is so much slower than USB? Should I look at
the card, the drive or the drivers? I would really like to use the
FireWire side of things, but not at this ridiculous speed.
With {probably} a different drive and dual interface caddy, my brother
had seen similar results ie usb2 is about 2x the speed of firewire. He
did notice that it is important to plug the second usb power header
connector in or either interface could be much slower than expected. I
think this might be because one interface /the drive might have been
running close to minimum power limits. {his test was with winxp sp2 on
a dell notebook}
The second thing to notice is that USB2 is a much newer standard. The
original "firewire is faster" comes from a comparison with USB1.
Could you try another comparison by writing a single large file eg a
dvd.iso to the drive ? I'd like to see the difference. Does it change
much if you blank the drive, and then test via the firewire interface
first.
Do you leave both interface cables plugged into the server when you
are doing the test ? This might cause an extra slowdown.
Also, for reference, what is the size of your previous test {du -s
whatever} ? And what did the write data-rate work out to be ?
It might be that a pure firewire interface would be better than the
dual interface box.
In any case, for backups you probably want to use something like:
$ rsync -a /home/db/pm/pm_10.d2 /media/extd/xxx1
DaveT.
