So i have this (relatively) old 450MHZ IBM Aptiva equipped with two NIC's. No high-end server, but i'm just playing around learning. However i'm only getting between 4.8 and 5.5 MB/s on my 10Mbit network. The other servers are playing just fine, but this one apparently has an issue.

I've used to bridge the NIC's to shape outgoing traffic. But after i realized the speed was about 40% of what it's suppose to, i've tested both NIC's separately, in and out, but no cigar.

Almost new Debian 4.0 install. Both NIC's are new and i've used them both in another server with good results.

Also, when i have bridge the NIC's, the first one, eth0, has a internet gateway. When i i.e. run ping google, it takes a long time for the box to react and start doing stuff. Maybe that's relevant.. shooting with my eyes closed here.. However, if i put down the second NIC it works "normally" again.

I realize this is one of those hopeless questions, but the problem is that i just don't know where to look (or start looking) for misconfiguration. I'll paste stuff and recheck the post often. Any ideas would be most welcome ;D

Joa

Do you mean you actually achieved a higher speed that 4-5 Mbit with a 10 Mbit NIC?

That is very unusual. AFAIK 4-5 Mbit is quite normal on 10 Mbit. Even when you go to 100 Mbit cards, your speed will be only around 20 Mbit tops.

The only applications where I ever have seen speeds which were exactly equal to the limit is with Internet connections. But there the speed is limited on purpose, and much lower than limits on hardware or protocol overhead.

jlinkels

I'm sorry, this wasn't really a good start.

First thing is that of course my network is 100Mbit, not 10. Would be silly to be getting 5MB/s on a 10Mbit NIC, yeah.

But i do get up to 11-12MB/s on parts of the network.. And this Debian server here is on that part.

So let's start over, shall we? ;D

What are you using to test/gauge bandwidth?

fyi, in my experience plain vanilla ftp xfers tend to be limited only by capacity of network while smb/cifs xfers tend to max out arbitrarily low and encrypted/scp/sftp xfers tend to be bottlenecked by cpu- sometimes approaching 40% of network capacity on slower systems.

if you're using 10/100 network equipment realistically expect 10-11MB/s max while 10mbit equipment may only reach 1-2MB/s.

If you are using 10/100 gear from end to end, the actual bottleneck on your network may actually turn out to be the system's disk controller (or rather the way the installed OS is interacting with the controller and/or storage hardware). Try hdparm -t -T /dev/XXX to reveal the max throughput of the drives on the system itself.

Hmm, this post really helped. It wasn't the disks, but the weak cpu that made it bottleneck. Although it doesn't choke the cpu entirely the secure programs transfers with alot lower speeds.

Thanks for the help ;D