Auto Port Aggregation behaviour

[Enrico Venturi]

Hello colleagues,
in our system we use a proprietary OSI stack wich directly connects to the LAN port by using the NM Id (or PPA);
We'd like to use the APA (Auto Port Aggregation) software to have LAN card redundancy at least from the IP standpoint;
my question is:
does the APA provide PPA virtualization (one PPA / NMId for two interfaces) or when a fault occur I have to redirect the stack to the new active board?

thanks
Enrico

[Alzhy]

Enrico,

It has its own PPA for the pair and technically you should not do redirection as APA will handle it nicely.

Here's a typical lanscan on a dual NIC'd server we have on APA:

Hardware Station Crd Hdw Net-Interface NM MAC HP-DLPI DLPI
Path Address In# State NamePPA ID Type Support Mjr#
LinkAgg0 0x0010837BFCF8 900 UP lan900 snap900 4 ETHER Yes 119
LinkAgg1 0x000000000000 901 DOWN lan901 snap901 5 ETHER Yes 119

[Enrico Venturi]

Nelson,
you mean that there isn't any relation between the active board and the PPA ...
or also, the APA represents two or more boards and when one of them fails this isn't manifest at PPA level ..
APA has its own PPA and I can refer the group of board simply by the APA PPA, isn't it?

[Alzhy · Accepted solution]

I think so. We've tried several times pulling a cable and the PPA stayed the same.

[cxtwo]

Hello, you should not have to redirect the stack to the new active board. APA creates a logical interface with PPA 1xx on HPUX 11.0 and 9xx on HPUX 11.11. This PPA replaces the 'true' PPA of the seperate interfaces.

You'll notice that if you do not have APA setup, you will see say, lan0 and lan1 in your lanscan output (PPA 0 and 1) but if the APA aggregate is up, then e.g. only lan900 is seen (PPA 900)

If one of the interfaces goes down, and fails over to the standby link, the PPA will still remain the same, so it should not affect your application.

hope this helps

[rick jones]

One thing to consider - the packet scheduling algorithms. You mention that you run an OSI stack. You also mention "from the IP standpoint" but I presume you mean an OSI Layer 3 protocol not "IP."

Anyhow, if you are running OSI, I suspect that means you can only have MAC-level packet scheduling because APA has not (to the bect of my knowledge, double check the manuals on docs.hp.com) been taught about OSI protocols.

If your goal was to have simple active/standby link failover you should still be in good shape.

If you wanted active/active, I _suspect_ you would only have outbound traffic spread based on destination MAC, which means if you were only talking to one other MAC on the network there wouldn't be much distribution.