|
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Aurélien Campéas wrote: |> Now before people start saying silly things, first that I have 2 |> other debian machines, but didn't bother buying network cards for |> them yet, as > there just desktops. | | | That's just plain silly :)) why mention these machines then ? | What I meant by this is that I wanted to avoid people suggesting that the problem was caused by purchasing linksys, or the like. |> If I run: ping -f -s 60000 172.27.2.27 from the three machines |> and target my Windows machine, the other machine with a 1 Gbps |> card, I can get about 45%, usage out of the network, according to |> windows, so the linksys switch seems to be able to handle it. If |> I run that just from the machine that has the 1 Gbps card, |> windows says its using about 20%, > which is the same as if I |> just run the command from the other debian machine, thats alot |> faster, but only has a 100 Mbps card. Futhermore if > I ping |> flood from my the 1 Gbps box, and then use the other debian |> machine to flood the 1 Gbps box, the traffic seems to cancel out. |> | | | I question the accuracy of your ping command. Let me suggest the | following scenario : | | apt-get install iftop vsftpd | | prepare some very big file on the debian machine and configure | vsftp such as to allow logging in from windoze | | launch iftop on a terminal | | download your big file from windows (with filezilla for instance) | and watch the iftop output ... | | Maybe that will help. I tried something similar, since my debian box already has a server on it, I tried using Download Accelerator and stream two files about 700 MB each. Download Accelerator then opened 7 connections per file, to my server and started downloading. The sum of the two connections maxed out at about 12000 KB/s, its get a more accurate number cause they would fight over them, but it seems to be around there. I think that ping is a good indicator of traffic because it basically sends packets as fast as it can. Although I wouldn't recommend it for actually getting a perfect measure of speed, in order to try and max out the network I would think that this would be a good indicator. Peter C. Norton wrote: On Thu, Apr 14, 2005 at 04:13:42AM -0700, Steve Ramage wrote: | mii-tools says that I'm running at 100 Mbps, but the man pages say | that it doesn't support 1000 Mbps (well actually it says it only | supports... and 1000 Mbps isn't there). I thought mii-tool was depricated. Stick with ethtool. | Furthermore ethtool seems to be just as useless as its information | seems to be completely irrelevant. For instance if I use ethtool to | change the speed, it doesn't. If I tell mii-tool to change it to 10 | Mbps, it does change it, but ethtool still says the linkspeed is | 1000 Mbps. If you're using gigE, there's no support in the standard for forcing the setting. You need to set the NIC to autoneg with ethtool, or else you may be setting it lower and confusing your dumb switch because it won't know about how to handle ports that are set to anything but autoneg. | Someone had suggested it might be the cable, but its not for a few | reasons. One the windows box has a Cat5 not Cat5e/6 cable and I | manage to crack 100 Mbps, and secondly, I dug up a Cat5e and it has | had no effect. Stick with the best cabling you can find. You're in a bind because if your switch is unmanaged you can't tell if the switch is seeing dropped packets or any other problem. Peter C. Norton wrote: | On Thu, Apr 14, 2005 at 04:13:42AM -0700, Steve Ramage wrote: | |> mii-tools says that I'm running at 100 Mbps, but the man pages |> say that it doesn't support 1000 Mbps (well actually it says it |> only supports... and 1000 Mbps isn't there). | | | I thought mii-tool was depricated. Stick with ethtool. ethtool doesn't seem to have any effect on my card, nor does it seem to be able to give my any useful information about it, ie, if I set it to 10 Mbps with mii-tool it still reports the speed as 1000, even tho everything else will say that its 10 Mbps. |> Someone had suggested it might be the cable, but its not for a |> few reasons. One the windows box has a Cat5 not Cat5e/6 cable and |> I manage to crack 100 Mbps, and secondly, I dug up a Cat5e and it |> has had no effect. | | | Stick with the best cabling you can find. You're in a bind because | if your switch is unmanaged you can't tell if the switch is seeing | dropped packets or any other problem. I don't actually believe that it is the cabling for a few reasons, first it's a Cat5e cable capping out at 12 MB/s. I have a Cat5 cable that goes twice the distance to my windows box, which has an onboard Gbps nic, that can go faster than 12 MB/s. | Since you have two systems, why not use netperf? Only one of them is a debian machine, and my options are limited on a Windows box, I might actually try knoppix in a bit. Roberto C. Sanchez wrote: | Have you considered that with a low end 900 MHz and slow FSB (133 | MHz) that you are going to have a hard time filling a Gigabit pipe? | Think about it. Gigabit represents ~125 MB/s. I have a similar | low-spec machine and have a hard time getting more than ~70 MB/s | across the PCI bus. Judd Tracy wrote: | You also have to remember that you only have 32bit pci, so you will | never acheive anything close to full speed with your current | motherboard. I should still be able to get more than 12 MB/s, PCI should max out at 133 MB/s. I personally would be happy if I could get 30 MB/s out of my server, I can always upgrade later. -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.2.5 (MingW32) Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org iD8DBQFCXvMFi3D5QAKwExARAlzWAJ995z/TGE/3XlV3J5BZ4VHBn3T0uACcCWEq /tcac9pn9FuX3NacrwCked8= =zcGg -----END PGP SIGNATURE----- |