Gateway probes

[Charles Keenan]

A relatively simple question ...

I have a two clients on two separate networks connected via two routers (HPUX mulit-homed servers).

On the clients I have configured static routes for both paths through the network. In my humble opinion a static rute is exactly that ... static ... We will use the first static route until something changes ...

I tried a test ... client1 routes through server1 first ... I pull the cable from the server to switch. As expected netstat -in shows the interface has a problem and we have lost the route to that network on the server. Here's my problem ... the client1 has two (static) routes to the distant network.
After 3 minutes + 10000 milliseconds shoould a route not become "DEAD" due to ip_rte_gw_probe=1 and ip_re_gw_prebe_interval=180000 ??? After +5 minutes the route is alive and well; even though he cable has been removed!!!

My memory seems to think the static route will remain but I need some clarification.

All machines are running HPUX 11i v1

[Ron Kinner]

You can see the Dead Gateways at the bottom of the display from

ndd -get /dev/ip ip_ire_hash

If it is listed there then it should not be used. It may not fall out of the netstat -rn table like you would expect. One of the following might have somehting to do with when they leave the table:

ip_ire_cleanup_interval:

Sets the time-out interval for purging routing table entries.
All entries unused for this period of time are deleted.
[5000, - ] Default: 300000 (5 minutes)

ip_ire_flush_interval:

All routing table entries are deleted after this amount
of time; even those which have been recently used. [60000, -]
Default: 1200000 (20 minutes)

Ron

[Charles Keenan]

Thanks Ron for that ...
I was looking at the routing table with ndd ( I was using ip_ire_status as well). The funny thing is the "type" of the routes added by the route command; they are seen as IRE_ROUTE. The "flag" field showed GW until I pulled the cable out and then it showed nothing. The problem (I think I am right) is that ip_ire_gw_probe only works for IRE_GATEWAY entries which are normally managed by gated. That will mean ip_ire_flush will be perfectly OK as gated will be updating the routing table on a regular basis ...

Can you confirm my theory ???

Mucho-ta

Charles K

[Ron Kinner]

Can't confrm your theory. Have seen a lot of complaints about the dead gateway probe causing problems with systems which do not run gated.

I think the real test would be to do a traceroute with the cable connected. Verify that it uses the gateway you think it should. Now pull the cable, wait for ndd to show it in the dead gateway list and repeat the traceroute and see if it uses the other gateway.

Ron