P4P800 Redhat 9.0 LAN Problems

Mogh · 07-17-2003, 05:07 PM

With some trouble I did get the 3COM adapter working using the drivers on the Asus CD and also the latest drivers from 06/03. However something is not right since I keep getting many TCP connection resets or refused when browsing the web, so there is only a handful of sites I can get into to.

The odd thing is I can ping those sites through DNS or IP just fine. At first I thought it might be an MTU issue but I adjusted that and continue to have TCP connection problems. I checked my router logs and I see it is loaded with connection resets from the P4P800 workstation.

I spent so much time on this I am ready just to put in an Intel adapter just to get it over with.

Any ideas?

**pointreyes** · 07-17-2003, 05:17 PM

Logged in as su what do you get when you type '/sbin/ifconfig eth0'?

Mogh · 07-17-2003, 06:13 PM

Everything looks good, no errors, ip information is correct. I tried changing my gateway to other routers on our network and it seems to work fine.

For some reason if my default gateway is this one particular router the tcp session resets going through it. The router log is filled with resets from the P4P800 station. Most of my stations go through that router do not have that problem but they have intel NICs. Very odd.

**SKI** · 07-17-2003, 06:52 PM

Are you running the same Kernel on the other workstations? This might have nothing to do with the NIC. There've been reports of some Kernels interpreting icmp-port-unreachable as tcp-reset. Problems appear to be related to elements such as Shorewall use and/or workstation firewall configurations. I know there is a Shorewall fix out there for this.

You might get a better handle on this by:

1) Increasing the size (to near MTU) & number of your ping packets, along with setting the DNF bit, to better simulate live traffic so you can observe behavior (if you haven't already done this).
2) Packet sniffing and analysis of comms between the routers and the P4P800.

...if you're motivated to dig into the prob. Otherwise, what the heck, drop in that NIC (assuming you have a spare on hand) and see if the prob goes away.

Mogh · 07-17-2003, 11:24 PM

The others are Linux RH 8. Considering I ran into a dozen other stability and bugs with RH 9 on this P4P800 I just got fed up and replaced it from scratch with RH 8. Solved all my other problems but this one.

Reducing the MTU did improve it some but still not quite there yet. I think there is likely some issue with the driver and will likely need to check the other things you mention.

**SKI** · 07-17-2003, 11:27 PM

What's your WAN type and current MTU setting?

Mogh · 07-18-2003, 12:54 AM

I adjust it to 1492 which would match the mtu on the router which is its default gateway.
TCP only resets when browsing sites while ping requests I cannot get it to fail.

The other router MTU is 1500 (no pppoe) and I can get across that router to sites just fine when I switch my default gateway to it.

Must be something in the driver as the other RH 8 workstations with intel adapters using the default mtu of 1500 have no problems getting across the pppoe router with the 1492 mtu.

Is there anything like mtu path discovery option that would be default for Intel adapters that the 3com on the P4P800 does not have?

**SKI** · 07-18-2003, 02:47 AM

This MIGHT be an issue if PMTUD is blocked on that particular router... someone would have to have blocked ICMP type 3, but leave type 0 and 8 open for this to happen and still allow ping. If your oversized pings are causing you to get "Packet needs to be fragmented but DF set" messages, you've proved that the basis of the PMTUD comms is working. Something else is going on here... Does there happen to be a firewall on that "one router"?

I'm going to nose around in my RedHat 7.3 install (I don't have 8 loaded at the moment) and see what MTU and MSS looks like. I'm running a Cisco router with MTU 1492/MSS 1452 (Std. PPPoE settings) and firewall featureset... maybe I can suss something out.

**SKI** · 07-18-2003, 03:40 AM

I'm running the straight-up defaults (MTU 1500) on my install with a crappy Linksys NIC... nothing to be seen there. Checked ifcfg-eth0 in "network-scripts" too, just in case.

Here's something to try...

"ping -c 4 -s 1472 www.cisco.com"

You should get a reply from that "1492" router: "Frag needed and DF set (mtu 1492)". Using "-s 1464" the pings should go through... I'm beginning to think that something (Driver??) has set DF for ALL comms, not just ICMP (echo request/ping) If you can ping @ 1464 reliably that pretty much confirms it. I've been thinking about it, and I believe that if DF is always set PMTUD will be defeated. The only work-around I can see (assuming I'm correct

) would be to lower MTU until you get reliable browsing or replace the NIC, and maybe post a bug-report if you want.

Mogh · 07-18-2003, 12:13 PM

I can ping reliably at 1464 as it should but no matter what I set the MTU to it just will not work
when web browsing. I just put an Intel Etherexpress Pro100b in the same workstation and it works fine without even touching the mtu.

Do you think it is the driver or a defective built-in 3com card in the P4P800?

I guess my next step is to put win2k on it and see if the same thing happens with the Windows driver. At least then I will know for sure whether it is the nic or the driver.

Mogh · 07-18-2003, 05:35 PM

I put Win2k on the P4P800 deluxe and it is working fine across the router. It also works fine under Redhat 8 using an Intel adapter.

So it looks like the problem exists when using the 3com drivers in Linux while browsing over a link that has a lower than 1500 MTU. Changing the MTU setting has no effect on correcting the problem.

Crap, as I wrote this, my firewall logs came back with some tcp resets from P4P800 station. It just looks like Win2k is more forgiving as I can still access websites despite all the resets.

Now all I have to do is find a email address to send this info to Asus in hopes of getting a new driver fix.

**SKI** · 07-18-2003, 10:49 PM

RMA the mobo if you can... If you're seeing similar behavior in win2k on the 3com the hardware is suspect. I have a hard time believing that 3com would not have a good driver for win2k. See what your supplier's tech support & customer service has to say.

Mogh · 07-18-2003, 11:16 PM

I am not so sure it is the board yet since I am finding other posts on the internet that describe the exact same problem in Linux.

From what I have read, have not tried it yet, the new kernels supposed to resolve to problem (i think 2.4.21 and above) or apply a sysconnect driver which is uses the same NIC.

Since I can browse the web in Win2k with relatively minor problems I likely would have missed it if it was not for the problems in Linux.

**SKI** · 07-18-2003, 11:39 PM

spagnitz · 07-23-2003, 10:00 PM

Quote:

Originally posted by Mogh
I am not so sure it is the board yet since I am finding other posts on the internet that describe the exact same problem in Linux.

From what I have read, have not tried it yet, the new kernels supposed to resolve to problem (i think 2.4.21 and above) or apply a sysconnect driver which is uses the same NIC.

Since I can browse the web in Win2k with relatively minor problems I likely would have missed it if it was not for the problems in Linux.

I have this problem also with the p4c800 (has the same onboard lan).

I have also heard this type of problem reported by others. One guy on usenet had this problem and then put in an intel pci lan card as the only change, the problem went away. I think there is some problem with the 3com drivers .

Also messed with the MTU etc.. like you have, couldn't get an improvement, Alos like you works fine on win xp (dual boot).

Kernel 2.4.21 does not help, I am also experimenting with kernel 2.6 but I haven't been able to get the drivers to compile properly yet.

Whats a "sysconnect driver"?

07-17-2003, 05:07 PM	#1
Mogh Registered User Join Date: Jul 2003 Posts: 39	P4P800 Redhat 9.0 LAN Problems With some trouble I did get the 3COM adapter working using the drivers on the Asus CD and also the latest drivers from 06/03. However something is not right since I keep getting many TCP connection resets or refused when browsing the web, so there is only a handful of sites I can get into to. The odd thing is I can ping those sites through DNS or IP just fine. At first I thought it might be an MTU issue but I adjusted that and continue to have TCP connection problems. I checked my router logs and I see it is loaded with connection resets from the P4P800 workstation. I spent so much time on this I am ready just to put in an Intel adapter just to get it over with. Any ideas?
(Offline)

07-17-2003, 05:17 PM	#2
pointreyes The race for quality has no finish line- so technically, it's more like a death march. Join Date: Feb 2001 Posts: 18,159	Logged in as su what do you get when you type '/sbin/ifconfig eth0'? __________________
(Offline)

07-17-2003, 06:13 PM	#3
Mogh Registered User Join Date: Jul 2003 Posts: 39	Everything looks good, no errors, ip information is correct. I tried changing my gateway to other routers on our network and it seems to work fine. For some reason if my default gateway is this one particular router the tcp session resets going through it. The router log is filled with resets from the P4P800 station. Most of my stations go through that router do not have that problem but they have intel NICs. Very odd. Last edited by Mogh : 07-17-2003 at 06:21 PM.
(Offline)

07-17-2003, 06:52 PM	#4
SKI Comms Moderator Join Date: Apr 2001 Location: Tempe, AZ (or wherever my luggage is) Posts: 8,487	Are you running the same Kernel on the other workstations? This might have nothing to do with the NIC. There've been reports of some Kernels interpreting icmp-port-unreachable as tcp-reset. Problems appear to be related to elements such as Shorewall use and/or workstation firewall configurations. I know there is a Shorewall fix out there for this. You might get a better handle on this by: 1) Increasing the size (to near MTU) & number of your ping packets, along with setting the DNF bit, to better simulate live traffic so you can observe behavior (if you haven't already done this). 2) Packet sniffing and analysis of comms between the routers and the P4P800. ...if you're motivated to dig into the prob. Otherwise, what the heck, drop in that NIC (assuming you have a spare on hand) and see if the prob goes away. __________________ Never try to teach a pig to sing... It wastes your time, and annoys the pig.
(Offline)

07-17-2003, 11:24 PM	#5
Mogh Registered User Join Date: Jul 2003 Posts: 39	The others are Linux RH 8. Considering I ran into a dozen other stability and bugs with RH 9 on this P4P800 I just got fed up and replaced it from scratch with RH 8. Solved all my other problems but this one. Reducing the MTU did improve it some but still not quite there yet. I think there is likely some issue with the driver and will likely need to check the other things you mention.
(Offline)

07-17-2003, 11:27 PM	#6
SKI Comms Moderator Join Date: Apr 2001 Location: Tempe, AZ (or wherever my luggage is) Posts: 8,487	What's your WAN type and current MTU setting? __________________ Never try to teach a pig to sing... It wastes your time, and annoys the pig.
(Offline)

07-18-2003, 12:54 AM	#7
Mogh Registered User Join Date: Jul 2003 Posts: 39	I adjust it to 1492 which would match the mtu on the router which is its default gateway. TCP only resets when browsing sites while ping requests I cannot get it to fail. The other router MTU is 1500 (no pppoe) and I can get across that router to sites just fine when I switch my default gateway to it. Must be something in the driver as the other RH 8 workstations with intel adapters using the default mtu of 1500 have no problems getting across the pppoe router with the 1492 mtu. Is there anything like mtu path discovery option that would be default for Intel adapters that the 3com on the P4P800 does not have? Last edited by Mogh : 07-18-2003 at 01:47 AM.
(Offline)

07-18-2003, 02:47 AM	#8
SKI Comms Moderator Join Date: Apr 2001 Location: Tempe, AZ (or wherever my luggage is) Posts: 8,487	This MIGHT be an issue if PMTUD is blocked on that particular router... someone would have to have blocked ICMP type 3, but leave type 0 and 8 open for this to happen and still allow ping. If your oversized pings are causing you to get "Packet needs to be fragmented but DF set" messages, you've proved that the basis of the PMTUD comms is working. Something else is going on here... Does there happen to be a firewall on that "one router"? I'm going to nose around in my RedHat 7.3 install (I don't have 8 loaded at the moment) and see what MTU and MSS looks like. I'm running a Cisco router with MTU 1492/MSS 1452 (Std. PPPoE settings) and firewall featureset... maybe I can suss something out. __________________ Never try to teach a pig to sing... It wastes your time, and annoys the pig.
(Offline)

07-18-2003, 03:40 AM	#9
SKI Comms Moderator Join Date: Apr 2001 Location: Tempe, AZ (or wherever my luggage is) Posts: 8,487	I'm running the straight-up defaults (MTU 1500) on my install with a crappy Linksys NIC... nothing to be seen there. Checked ifcfg-eth0 in "network-scripts" too, just in case. Here's something to try... "ping -c 4 -s 1472 www.cisco.com" You should get a reply from that "1492" router: "Frag needed and DF set (mtu 1492)". Using "-s 1464" the pings should go through... I'm beginning to think that something (Driver??) has set DF for ALL comms, not just ICMP (echo request/ping) If you can ping @ 1464 reliably that pretty much confirms it. I've been thinking about it, and I believe that if DF is always set PMTUD will be defeated. The only work-around I can see (assuming I'm correct ) would be to lower MTU until you get reliable browsing or replace the NIC, and maybe post a bug-report if you want. __________________ Never try to teach a pig to sing... It wastes your time, and annoys the pig.
(Offline)

07-18-2003, 12:13 PM	#10
Mogh Registered User Join Date: Jul 2003 Posts: 39	I can ping reliably at 1464 as it should but no matter what I set the MTU to it just will not work when web browsing. I just put an Intel Etherexpress Pro100b in the same workstation and it works fine without even touching the mtu. Do you think it is the driver or a defective built-in 3com card in the P4P800? I guess my next step is to put win2k on it and see if the same thing happens with the Windows driver. At least then I will know for sure whether it is the nic or the driver.
(Offline)

07-18-2003, 05:35 PM	#11
Mogh Registered User Join Date: Jul 2003 Posts: 39	I put Win2k on the P4P800 deluxe and it is working fine across the router. It also works fine under Redhat 8 using an Intel adapter. So it looks like the problem exists when using the 3com drivers in Linux while browsing over a link that has a lower than 1500 MTU. Changing the MTU setting has no effect on correcting the problem. Crap, as I wrote this, my firewall logs came back with some tcp resets from P4P800 station. It just looks like Win2k is more forgiving as I can still access websites despite all the resets. Now all I have to do is find a email address to send this info to Asus in hopes of getting a new driver fix. Last edited by Mogh : 07-18-2003 at 05:45 PM.
(Offline)

07-18-2003, 10:49 PM	#12
SKI Comms Moderator Join Date: Apr 2001 Location: Tempe, AZ (or wherever my luggage is) Posts: 8,487	RMA the mobo if you can... If you're seeing similar behavior in win2k on the 3com the hardware is suspect. I have a hard time believing that 3com would not have a good driver for win2k. See what your supplier's tech support & customer service has to say. __________________ Never try to teach a pig to sing... It wastes your time, and annoys the pig.
(Offline)

07-18-2003, 11:16 PM	#13
Mogh Registered User Join Date: Jul 2003 Posts: 39	I am not so sure it is the board yet since I am finding other posts on the internet that describe the exact same problem in Linux. From what I have read, have not tried it yet, the new kernels supposed to resolve to problem (i think 2.4.21 and above) or apply a sysconnect driver which is uses the same NIC. Since I can browse the web in Win2k with relatively minor problems I likely would have missed it if it was not for the problems in Linux.
(Offline)

07-18-2003, 11:39 PM	#14
SKI Comms Moderator Join Date: Apr 2001 Location: Tempe, AZ (or wherever my luggage is) Posts: 8,487	__________________ Never try to teach a pig to sing... It wastes your time, and annoys the pig.
(Offline)

Thread Tools
Show Printable Version Email this Page
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode