How does APA and ServiceGuard work together

[Mike Keys]

We will be implementing two rp4440's with VA7110 shared between the nodes. The nodes will run through two brocade switches as part of the HA no single-point of failure config.

I was told today that each server is required to have a minimum of 3 NICs on each node for HA. One primary, one for heartbeat and one for failover. On top of this, we will also be implmenting HP APA.

My question is how will this work. Does MC ServiceGuard control of the NIC failure between the primary and fialover card or does APA or both?

[Steven E. Protter]

APA will let you do a few things:

1) Use both ports on a multi-port NIC as the same ip address, doubling bandwidth and improving reliability.
2) Team multiple NIC cards with the same IP address, improving reliability and throughput.

It will work with serviceguard, but may increase your NIC requirements further.

SG recommends the heartbeat lan be a private lan. You wold not like your nodes to TOC because normal network congestion cuts off the heartbeat for 20 or 30 seconds.

It would seem that things are being a little over-engineered. Still the standby lan is easy, because its down and won't interfere with other NIC operations.

Note: with the exception of APA NIC cards, you can never have two NIC cards up on the same network on hp-ux. This will crash networking. Take it from one who has done it.

Back to your last question. SG does not handle reliability with the APA NIC's. APA does. SG will work with APA and if the whole APA NIC group fails take that IP address to a standby LAN card if configured to do so.

SEP

[Mike Keys]

O.K. let's try this scenario.

We currently have one server on 129.47.14.14 and one on 129.47.8.8.

We implement SG.

Therefore, on the 129.47.14.14 we have 1 NIC with that IP, one NIC on a different dedicated sub for the heartbeat and one not configured for failover. The same config for node 2 except that it's primary NIC has 129.47.8.8

The question is how do I configure that standby NIC. With APA, I would run in LAN_MONITOR more and have this card as a standby. But, for SG this card also needs to assume the role of the 129.47.8.8 node if that node has a problem. And you say that I can't have two NIC's up on the same machine without crashing the network.

O.K. so let me see. We can change node 2 to run on a different sub. Not a problem. But still, how do I set-up the failover card. Do I config with APA and leave in LAN_MONITOR mode and then SG has some sort of config as well to also make that card the failover card for node 2?

Or, do I need to have one additional NIC for every NIC card that I want to provide failover for using APA?

[Patrick Wallek]

If all you want is failover in case one NIC on a server goes bad, then use either APA or MC/SG, but not both.

If you try to do failover with both APA and MC/SG I think things are going to get very confused.

Just configure MC/SG to use the 3rd card as the standby NIC for your production LAN and all will be well.

[Sridhar Bhaskarla]

Mike,

I just didn't understand your question. If you are going to have three interfaces configured with three different IP/subnets on each server, then where are you going to implement APA?. I assume that you are going to aggregate multiple NICs for each IP. In that case, simply specify the APA interfaces (say lan100) in ServiceGuard configuration. You will not have any 'standby' configuration as APA will take care of availability also.

-Sri

[G. Vrijhoeven]

Hi,

Lanmonitor can not be used in combination with fail over from MC/SG. It is possible to use a trunk using an APA interface in MC/SG. For the cluster config it is just a lan interface (e.g. lan100)

To enable a fail over interface make sure the failover interface is patched in the same vlan on a (if it is possible) different switch.
Adjust the cluster ascii file and do a cmcheckconf and a cmapplyconf.

HTH,

Gideon

[Mike Keys]

I don't want to make this harder than it is.

Is it possible to use APA and Service Guard together?

If so, how do they interact with one another?

Who assumes failover responsibility between nodes (node failure, loss of heartbeat) and if the primary NIC goes down on a particular node?

If Service Guard, then how is APA configured to work with it?

What mode should APA running in to work with Service Guard?

Is there an example configuration that I can look at or read about?

How many NICs would I need to use APA and Service Guard together?

[John Poff]

Hi,

We've used APA and MC/SG together here before. They work just fine together. First, you configure and get APA working by itself. APA will present you with one or more virtual LAN interfaces that will look like lan100, lan101, etc. You just plug those LAN interfaces into your MC/SG config (instead of lan1, lan2, etc.)

For a failed card, APA handles it. MC/SG doesn't know the difference. Let's say you have three LAN cards ganged together using APA, which presents them as lan100. One of the LAN cards fails. You will still have lan100 active and working, with just two cards now. MC/SG doesn't see anything different, so it doesn't care.

Now, for your heartbeat, you still probably want to configure a separate heartbeat LAN as others have mentioned. Usually the boxes we use for nodes in our MC/SG cluster have built-in LAN interfaces, so we just get a small hub and configure them to use a separate little subnet just for heartbeat traffic.

As for the modes, I don't remember off the top of my head since we don't run APA any more. Here is a link to the APA documentation:

http://docs.hp.com/hpux/netcom/index.html#Auto%20Port%20Aggregation%20%28APA%29

It covers using APA and MC/SG together.

Good luck!

JP