Flex-10/FlexNIC and MAC addresses

[eran70]

We start to use Flex-10 VC modules and BL460 G6 servers
We successfully use FlexNIC - we see 8 NIC's per server in Virtual connect and in Windows, and we can allocate bandwidth for NIC in VC manager
Question:
I just see Jarkko K. answer here:
http://forums13.itrc.hp.com/service/forums/questionanswer.do?threadId=1350456
and I understand from the answer that Virtual connect defined MAC addresses must be used for Flex-10/FlexNIC
We use static, factory-default MAC addresses in VC domain and it's work too.
So, question is: is it really required to use Virtual connect defined MAC addresses for Flex-10/FlexNIC using ? Is it's described in any manual/white paper ?
Thanks !

[Adrian Clint]

I would ************ HIGHLY RECOMMEND ************ that you use the Virtual Connect defined mac addresses.

There are scenarios when you have more than one enclosure with VC and you start to move blades between them where you can result in two servers having the same mac address on one or more NICs.
If you use VC defined and you use a different defined range for each enclosure then then this will not happen.

[Michael Leu]

I don't have Flex-10 yet, but from what I have read in the "Flex-10 technology brief"
http://h20000.www2.hp.com/bc/docs/support/SupportManual/c01608922/c01608922.pdf
it says on the top of page 13 that you should get 8 factory-assigned MAC addresses on a Flex-10 capable server.

[Adrian Clint]

Yes you get the 8 factory MAC addresses for each physical server. But its wether you use those or you use Virtual Connect predefined ones.
As mentioned above there is a simple scenario that results in two servers having the same MAC address (for one or more NICs) if you use the physical MACs. Therefore I recommend to customers that they do not use physical MACs but VC predefined MACs.

[Michael Leu]

Adrian, what I don't understand is how you can end up in this scenario when the factory-defined MACs are tied to the hardware?

[Adrian Clint]

You plug in server A & B into enclosure 1. MACs from both servers A&B are now in the VC MAC pool for enclosure 1.
You remove server A and plug it in with server C in enclosure 2. Now MACs from both A&C are now in the VC MAC pool for enclosure 2.
And if you have both enclosure 1 & 2 plugged into the same switch then you could have the same MAC on different servers connected to the same switch. Whoops!

If you stack the enclosures or use VCEM on both enclosures before the move then you will be able to avoid this problem.

[Michael Leu]

Thanks for the explanation Adrian. I was not aware that the factory MAC stays in the pool.

(And sorry about the thread highjacking ;-)

[eran70]

Ok, so it is only recommendation and not requirements

[The Brit]

For the OP,

A more compelling argument (which includes all of Adrians arguments as a subset) is that if you have a Server Blade fail (not unheard of, I'm told), then the VC MAC's (stored in the Blade Profile), allow the blade to be replaced with a completely new blade, without any need to mess with the network connectivity.

I personnally think this trumps all of the other considerations, (after all, it is what VC is all about!)

Dave.