Lan card starts to work then it stops working

[Pete Des Moulins]

We have an L3000 with two single 10/100MB btlan
cards. One is set to our backup network and continues to function normally, the other starts up when the system reboots and then drops out after about two minutes.
The network parameters are as follows:
IP 10.128.172.86
Subnet mask 255.255.255.0
BROADCAST 10.128.172.255

When the m/c reboots I can see the m/c over our
PC network using RUMBA for HP. After about two minutes the RUMBA session locks up and we can only see the rest of the 10.128.172.86 network from the console. I can ping other m/c's on this network from the console and database functions still work. The VLAN of this network is on the development side of our Firewall and the Firewall people can see packets going to & fro for this 86 address. THere are no error messages or error logs generated and the card exercises properly with STM. When I swap the cable to another BTlan card the same lock up occurs

[Shaikh Imran]

hi,
At this point
what is the o/p of ?
#ioscan -fnC lan
Is the other lan card status claimed ?

Regards,

[Pratyush Paul_1]

Hi -

Please check the speed, duplexity and autoneg (on/off). Probably you have a wrong setting in your switch/router. This happens when you do not have the auto neg turned on.
This speed & setting mismatch is creating the problem. Try to use ndd and see if there are some errors on your NIC. Do you have ICMP request turned off in your router( does not respond to ping router). Hope this helps. Let me know if you are able to ping your router.

regards

Pratyush Paul

[Thomas Bianco]

BTLan cards have been known to fail at autodetection silently. they usually fail to 100 half, where as the switch side (cisco at lease) will go 100 full.

we had lots of fun with older HPUX boxen falling off the network due to this.

[rick jones · Accepted solution]

You mention two NICs, but only one IP and subnetmask - I trust the two NICs are in separate subnets, or you have ip_strong_es_model set to 1?

I can see where a speed mismatch would cause a system to appear to drop off the net, but a duplex mismatch usually just means really bad performance. One sign of duplex mismatch would be seeing _late_ collisions in lanadmin statistics when the card is in half-duplex, or FCS errors when the card is in full-duplex.

Since you mention firewalls and such, will the firewall respond to ICMP Echo (aka ping) requests on the "inside" (I'm assuming your HP-UX system is on the inside). By default, HP-UX checks the health of routes (such as the default route) by pinging them. If no ping replies are received, the route is marked dead. If that is the only route to everyone else, then the system will appear to drop off the network. Setting ip_ire_gw_probe to 0 will disable the dead gateway detection.

To check that a NIC is alive and well at the link-level you can use linkloop - so long as there is another HP-UX system connected to the same VLAN or some other system that responds to 802.3 XID or TEST frames.

[Thomas Bianco]

your point is taken, i should have clairified my statement.

i should have said, "we had a lot of fun with the cisco switch killing the ports the HPUX boxen were attached to due to excessive errors and knocking the hpux boxes off the network."

[Pete Des Moulins]

Thanks to all the suggestions. Rick Jones hit it right on the nose. The gateway was a
Firewall in our testlab and it did not respond to pings. Thanks Rick

[rick jones]

glad to help - now if you could go swat the firewall (or the firewall admin :) on the nose with a rolled-up newspaper... :)