HP APA

[James Carbonaro]

Does anybody know how to get the PRIMARY link to failover in the event that there is still link status, but routing is not available? I have tried manipulating POLLING_INTERVAL and DEAD_COUNT, but all that seems to happen is that a nettl message is sent to the console. The problem is particular to Nortel switches where a single control board controls all cards in the chassis. If something goes wrong with the main controller, routing will cease, but other boards will maintain link up state. Seems to me that there needs to be some kind of triangulation algorithm where in addition to polling each other, the members of the linkagg group could poll for a third MAC address outside the group. Any ideas?

[Jim Keeble]

You are correct, HP APA does not provide this service for lan monitor failover groups.

You might try setting up each of 2 routers as default gateways, and using dead gateway detection as a means detecting failure. See the help for "ndd -get /dev/ip ip_ire_gw _probe" .

Also, Cisco has HSRP to make a failover transparent to nodes on the subnet, I don't know if Nortel has a similar function.

[rick jones]

APA is concerned only with link-level (aka MAC) connectivity - despite some of the packet schduling algorithms looking beyond the Ethernet header :)

It is not meant to deal with situations where an IP router is no longer among the living - I am ass-u-me-ing that when you say "routing" you mean IP routing and not MAC-layer switching.

Problems with routers going away can be dealt with in perhaps a few different ways:

*) use the Router Discovery Protocol Daemon - this will keep your "default" route refreshed based on rdp packets it does/does not see from your first-hop routers

*) might be able to configure a second router and let HP-UX dead gateway detection do its thing

*) go for broke and run gated and wiretap the routing messages

or did you simply mean that when the main board fails cross-board (?) switching ceases to function?

[James Carbonaro]

Rick,

To clarify, when the control board dies, all switching basically ceases. Since the standby port is connected to a different switch configured to be part of the same VLAN, it should be able to take over control, but this doesn't happen. What I observe is that a nettl message is sent indicating that status messages are not being received by the standby port, but since the primary still has a link up state, I also receive a nettl that status messages aren't being received by the primary port. Using lan analyzers connected in-line shows that the unavailable port continues to send out linkloop packets continuosly, but of course nothing can respond.

[rick jones]

I'm gong to get very rapidly out of my explicit PA depth here, but if I understood your message correctely, the switch becomes one of the undead. it no longer responds to any LACP messages on either port, but the link status remains up on each port as well.

It would seem - at least on the surface - that APA should failover on the OR of link-down and no control response. As such, I think you should file an official defect against APA.

I'm not sure that sending frames off a third MAC is really the way to go - that it seems is just adding another failure point if that third MAC simply went away for some other reason. Also, the protocols et all still need to work when there is just a back-to-back situation.