Problems with routing

[Michelle Alsop]

I have two systems that are experiencing odd routing behaviour. If I try to ping node spica directly it fails with network unreachable. However, if I ping its gateway I can then get through to node spica. This connectivity persists for a few minutes and if I go away and come back to it later the ping no longer works. The source node name is psprdap1 and the target is spica. I enclose the output from a script session for your reference

[Duncan Edmonstone]

Michelle,

That's a very strange routing table... any reason you're not using a default gateway?

Can you post the output of a traceroute to the gateway (10.253.254.1) cos looking at these routing tables I can't see how your pings are even getting there (unless I'm going blind!)

PS... think I know this site - have a vague recollection of putting a cluster into a warehouse there once about 6 years ago...

HTH

Duncan

[Srikanth Arunachalam]

Hi,

I think the router information is not configured appropriately. Please let us know what is the output of

traceroute spica

add appropriate router address on psprdap1 using

route add net 1

If router information is already present is wrong, make use of
route delete

If tcpip parameters are not set correctly make use of /usr/bin/ndd to set them correctly.

Thanks,
Srikanth

[Srikanth Arunachalam]

Hi,

I think the router information is not configured appropriately. Please let us know what is the output of

traceroute spica

add appropriate router address on psprdap1 using

route add net 1

If router information is already present and is wrong, make use of
route delete

If tcpip parameters are not set correctly make use of /usr/bin/ndd to set them correctly.

Thanks,
Srikanth

[Sandman!]

What is the subnet mask of spica or what network address is it a part of? Is it 10.253.0.0? If that is the case then routing table entry listed below is incorrect:

> 10.253.0.0/255.255.255.255 10.253.254.1 UGH 0 lan0 0 <

If spica belongs to the 10.253.0.0 network then remove the above host route i.e.

# route delete host 10.253.0.0 10.253.254.1 1

~cheers

[Hank Classe]

I don't know if this will help you but have a look.....

I have seen this issue first hand in a Cisco shop. In that case, the ARP timeout value of the Cisco switches (which also did routing) were not correctly tuned. What happened in that case was the HP-UX system aged out its arp entry before the the router/switch aged its entry out. The HP-UX system would flood the VLAN with an arp request but the router/switch would not respond with the arp reply since the entry was still in its arp table. Cisco TAC recommended adjusting the arp timeout value for all switches and routers supporting the VLAN to a lessor than the arp timeout of the UNIX & other servers on the effected VLAN to what happens to be their default value of 600 seconds. This resolved the issue, but only after the value had been properly adjusted on ALL switches and routers for the VLAN impacted by the problem.

[Michelle Alsop]

Thanks for all your help and advice. It seems that the proble has fixed itself after "networks" took a look at it.

Duncan you may have indeed put a cluster in at one of our sites. It still works

Thanks

[Michelle Alsop]

Closing thread as problem fixed itself but I am taking the advice of all who replied to resolve this strange routing