View previous topic :: View next topic |
Author |
Message |
pcassidy Tux's lil' helper
Joined: 19 Jul 2003 Posts: 116 Location: Dublin, Ireland
|
Posted: Fri Aug 15, 2003 1:03 pm Post subject: nforce2 network performance |
|
|
First Time poster, some time lurker, please bear with me
I'm noticing that the network performance using the nvnet module isn't all that it could be,
has anyone else noticed this?
at home I've got an abit NF7-M motherboard, on a quiet network with only 2 machines.
The other machine has a standalone network card.
I've tried all the options to the module, (optimization, speed, duplex) but they make little difference to transfer speeds.
As an experiment I tried transferring the source bz2 for openoffice(158M) across the network.
I tried this both at home. and also in work on a busy network with hundreds of users, I don't know what the motherboard is in work, but it's a DELL machine running RedHat 7.3 with a 3com network card.
comparison
time cp OOo_source.tar.bz2 /tmp on both networks.
At work transfer took 14 seconds, at home 36 seconds.
Does this seem slow to anyone?
Also is there anyway to query the nvnet driver. The other network card logs it's speed to /var/log/everything when it initiializes. The nvnet driver is silent to a fault.
Thanks,
Paul.
--
I don't believe in sigs. |
|
Back to top |
|
|
69link n00b
Joined: 18 Apr 2003 Posts: 53 Location: Sweden
|
Posted: Fri Aug 15, 2003 1:32 pm Post subject: |
|
|
How are you copying the file? FTP? Samba? NFS?
Try FTP instead if you are using something else and come back with the results. |
|
Back to top |
|
|
pcassidy Tux's lil' helper
Joined: 19 Jul 2003 Posts: 116 Location: Dublin, Ireland
|
Posted: Fri Aug 15, 2003 2:42 pm Post subject: |
|
|
Thanks for the response.
ftp at work gives.(tried it 3 times)
165475319 bytes received in 14.6 secs (1.1e+04 Kbytes/sec)
165475319 bytes received in 15 secs (1.1e+04 Kbytes/sec)
165475319 bytes received in 16.8 secs (9.6e+03 Kbytes/sec)
I'll post the results using the nforce2 board when I get home.
thanks,
Paul. |
|
Back to top |
|
|
pcassidy Tux's lil' helper
Joined: 19 Jul 2003 Posts: 116 Location: Dublin, Ireland
|
Posted: Fri Aug 15, 2003 4:53 pm Post subject: |
|
|
Finally, much emerging required (xinetd etc.)
from home times are
165475319 bytes received in 35.3 secs (4.6e+03 Kbytes/sec)
165475319 bytes received in 35.5 secs (4.5e+03 Kbytes/sec)
165475319 bytes received in 36.6 secs (4.4e+03 Kbytes/sec)
165475319 bytes received in 36.6 secs (4.4e+03 Kbytes/sec)
The fourth transfer was with a crossover cable just in case my hub was the problem.
So it looks like something is running slow.
Any clues?
thanks,
Paul |
|
Back to top |
|
|
69link n00b
Joined: 18 Apr 2003 Posts: 53 Location: Sweden
|
Posted: Fri Aug 15, 2003 5:14 pm Post subject: |
|
|
Hm, a tricky one...
Tell me what you get by running
# hdparm -tT /dev/hdXX
and by running
# mii-tool |
|
Back to top |
|
|
pcassidy Tux's lil' helper
Joined: 19 Jul 2003 Posts: 116 Location: Dublin, Ireland
|
Posted: Fri Aug 15, 2003 6:45 pm Post subject: |
|
|
I think you might be onto something.
On the receiving machine(with the Abit MB)
hdparm -tT /dev/hda
/dev/hda:
Timing buffer-cache reads: 1100 MB in 2.00 seconds = 550.00 MB/sec
Timing buffered disk reads: 142 MB in 3.03 seconds = 46.86 MB/sec
looks good to me. Checked this a couple of days ago as others with
the nforce chipset seem to have problems. DMA is on.
mii-tool
eth0: negotiated 100baseTx-FD, link ok ( thanks I was wondering how to query ethernet connections)
But on the Sending machine. A really old pentium 166
hdparm -tT /dev/hda
/dev/hda:
Timing buffer-cache reads: 128 MB in 2.96 seconds = 43.17 MB/sec
Timing buffered disk reads: 64 MB in 4.39 seconds = 14.56 MB/sec
mii-tool
eth0: negotiated 100baseTx-FD, link ok
Looks like the IDE (ATA-33 IIRC) of the old machine can't keep up
with the network i/f
Case closed?
In any case thanks for the help. |
|
Back to top |
|
|
Ox- Guru
Joined: 19 Jun 2003 Posts: 305
|
Posted: Sat Aug 16, 2003 4:07 am Post subject: |
|
|
It seems to me that might be the problem, especially if your work machines are in the 45 MB/sec range... that's 3 times faster disk, which corresponds to 3 times faster on the transfer that you were seeing.
Just as a general note, you shouldn't test NIC performance on 2 different networks or even on 2 different machines.
Your work network might have more traffic (possibly with packet collisions) and it might have one or more switches on the route (which will have different performance characteristics from a single home hub/switch).
If you test on 2 different machines you don't know what other factors (like disk performance) are affecting the benchmark. |
|
Back to top |
|
|
|