Internet connection slow, LAN connection fine

[tameboy]

the tg3 driver is the official broadcom supported driver. It would appear they expect you to use ethtool to configure it, so this is something I can look into.

I will also check IP and subnet addresses on other machines on the network ...

[tameboy]

Originally Posted by Hlingler

Hello tameboy:

Could be SELinux interfering with Smolt, but that can be sorted out later, since you have posted the output.

The netmask looks wrong - I thought that it should be 255.255.0.0 for Class B addresses (172.x.x.x). Where did you get this number or why did you make it so?

Regards,
V

The subnet mask and IP address ranges are those used by the existing windows machines on the network. They were set up by the company who run our main company software (on another server).

Since a couple of you have raised queries about the subnet mask and broadcast IP I have done a lot of reading regarding this and I can't see what is wrong with the values. These values work for all the other machines on the network (and used to work very well on this server in Fedora 7) I don't think they are likely to be at fault.

I've changed the router port with no success. I'll keep on with trying the other suggestions.

[krolaw]

I am experiencing the exact same problem. But I have something to add to the confusion. This machine acts as server for a school and part of its intended operation is transparent web proxy. Users were complaining the web was slow and it was. I thought the problem was with squid so I configured iptables to SNAT the data straight through (server ip 249):

Chain POSTROUTING (policy ACCEPT)
target prot opt source destination
SNAT all -- !192.168.1.249 0.0.0.0/0 to:192.168.1.249 # Straight through

Now, web access for clients is FAST even though it's going through the server. Yet if the server tries to download something it's limited to like 8k.

Summary:
Server as Gateway for clients: FAST UP, FAST DOWN
Server Own access to Internet: FAST UP, SLOW DOWN

Aside from downloading updates being impractical, I really need to get proxy services back online. This is weirdest network problem I have encountered. I mean, if server access is slow, shouldn't pass through be slow too?

Additional Info:
[root@www ~]# dmesg | grep eth0
udev: renamed network interface eth0 to eth1
e1000: eth0: e1000_probe: Intel(R) PRO/1000 Network Connection
e1000: eth0: e1000_watchdog: NIC Link is Up 100 Mbps Full Duplex, Flow Control: RX/TX
eth0: no IPv6 routers present

Any suggestions appreciated.

[tameboy]

Hi Krolaw

Sorry to hear you are having the same problem. I have tried all the suggestions in this thread to no avail. I have posted on the HP ProLiant ML115 forum which has proved helpful in the past, but so far no result there either. I too would be grateful for any more suggestions, and if you do find out what is causing your problem please let me know. It certainly sounds like it might be a similar issue.

[Hlingler]

@tameboy:

You are using a router to connect one of your PCs to the internet, correct? If so, what is the make and model of that router? Have you tried to access it, to configure any settings in it? My DSL router, for example, has a small mind of it's own, with all kinds of internal settings and stuff (accessible via web browser), anything could go wrong is possible. I know that you said everything was working before with F7, but maybe when you switched to F8, something in the router "burped".

V

[krolaw]

I'm beginning to seriously think that it's an incompatibility between router and FC8 (FC1 seemed fine). I have two locations with Allied Telesyn and FC8 and they are fine. It's the Cisco (with two servers 4 years apart in age, both with FC8) that has the problem. I'll probably end up swapping the Cisco out temporarily with something cheap just to test it.

Cheers.

[madopal]

krolaw, please post if the Cisco is the issue, and if so, if there's ANY way you can figure out what's causing the problem. I've got the same issue here at work. The connection speeds seem fine within our shop, but once I head outside, bad news. I have a WinXP box on my desk next to me that doesn't have the same problem, so it'd be interesting to find out what's in the packets that is confusing F8.

[krolaw]

Originally Posted by madopal

krolaw, please post if the Cisco is the issue...

I'll be temporarily swapping routers in the next 48 hours.

Will post here.

[krolaw]

Ok I swapped out the Cisco ADSL Router with an old cheap DSE ADSL router and my 2x Fedora Core 8 servers there had internet at full speed. At two other locations (also FC8) with Allied Telesyn ADSL routers, no problems either.

For those that have the same problem, what type of routers are you using? My problem appears Cisco related, or at least a problem with config on the router. Still, this was not an issue before moving to FC8 (previously FC1) and Windows/Mac (anything but FC8) machines are unaffected.

[madopal]

We have a Cisco 871 here. I can dig into it and see what the settings are that might be doing something. Although it is peculiar that only my linux boxen (going to try some live distros to see if others have the problem) seem to be slow going out.

Can anyone share their ethtool/mii-diag settings when it's failing to see if we've got something similar?

Here's what I have:

Settings for eth1:
Supported ports: [ TP ]
Supported link modes: 10baseT/Half 10baseT/Full
100baseT/Half 100baseT/Full
1000baseT/Full
Supports auto-negotiation: Yes
Advertised link modes: 10baseT/Half 10baseT/Full
100baseT/Half 100baseT/Full
1000baseT/Full
Advertised auto-negotiation: Yes
Speed: 100Mb/s
Duplex: Full
Port: Twisted Pair
PHYAD: 0
Transceiver: internal
Auto-negotiation: on
Supports Wake-on: pumbg
Wake-on: g
Current message level: 0x00000033 (51)
Link detected: yes

[gsgleason]

This could be your problem right there:
Advertised auto-negotiation: Yes
Speed: 10Mb/s
Duplex: Half

Your server is connected to a switch port that is likely not using auto negotiation, and is likely locked a certain speed and duplex. This can cause a duplex mismatch and causes problems when receiving large files.

Check the port settings for whatever it's connected to.

If the device that your server is connected to is a hub and not a switch, then 10/half is correct as hubs do not support full duplex operation since all ports are in the same collision domain.

The large number of collisions is likely due to your switch port operating in full duplex mode and trying to send data to your server while receiving data as well, which is correct for full-duplex operation. However, your nic running at half duplex considers this a collision (CSMA-CD), stops transmitting, and starts over, and the full dup port on the switch sees this as an error.

I see this ALL THE TIME in my line of work.

Again, see what port your server nic is connected to, and look at its settings. We won't know until then.

[krolaw]

Fedora 8 (like Vista - yuck), switches on TCP Windows Scaling by default. It just so happens that the Cisco 871 doesn't support this. Perhaps later Cisco firmware is needed. In any case, I disabled TWS and voila, full speed again.

edit /etc/sysctl.conf and add the following lines

#Control TCP Windows Scalling
net.ipv4.tcp_window_scaling = 0

http://en.wikipedia.org/wiki/TCP_window_scale_option
http://forums.fedoraforum.org/showthread.php?t=179814

Cheers.

[madopal]

Correction...that did work for me. Nice find, krolaw! Thanks!

[tameboy]

Originally Posted by krolaw

edit /etc/sysctl.conf and add the following lines

#Control TCP Windows Scalling
net.ipv4.tcp_window_scaling = 0

Guess what? We've got a Cisco router too. And your two little lines of code have worked wonders for my connection speed. Thanks a million Krolaw, brilliant work!

Now I've got to re connect yum to some online repos and get my system updated again.

Really can't thank you enough for that little gem.

[csonne]

Originally Posted by krolaw

I'm beginning to seriously think that it's an incompatibility between router and FC8 (FC1 seemed fine). I have two locations with Allied Telesyn and FC8 and they are fine. It's the Cisco (with two servers 4 years apart in age, both with FC8) that has the problem. I'll probably end up swapping the Cisco out temporarily with something cheap just to test it.

Cheers.

Hello Korlaw,
I have also the same problem with a cisco 836, i wil give it a trie tomorrow