On 2/18/19 5:02 PM, Bin XA Xu wrote:
Hi Thomas,
     Just a bit confusing about your context. So you have a node with 2 NICs are enabled PXE boot in BIOS, and both NICs are connected in the same network with MN's provision NIC?      If this is your case?  why we have the two NIC in the same LAN, do you want a bonding after provisioning?     Or the two NICs are connected into different networks, while MN's dhcp server listen on both the networks.  Could you brief describe your cluster networking design?

Hello, sure :

- the cluster is on a private non routed ethernet/ip network
- I've got only one MN with 2nics
    - one on a public routed network just for sysadmin to access it
    - one on the cluster private network
- nodes have
- 2 onboard 1Gb/s nics (eth0, eth1) eth0 is connected, eth1 is disconnected - 1 additional PCI-E 10Gbs/nic (eth2) eth2 is connected to a different switch than eth0 - we're using switch-based discovery mapping switch/port eth2 is connected to
- BMC is configured in shared mode and physically uses the eth0 nic
- before we had 10Gb/s connectivity, data (as opposed to ipmi traffic) was allowed through eth0 as well and we kept that switch configuration
- we do not use any bonding

So boot order in BIOS is typically configured like (or might be) this :

first : network boot
   XE (eth2)
   GE1 (eth0)
   GE2 (eth1)

node xCAT object has only one ip derived from a regexp in hosts table

So we PXE on eth2 and at the end of provisionning eth2 (10Gb/s nic) gets the ip.

If for some reason GE1 gets set to appear before XE, the node would be correctly discovered/provisionned (given it would have already been previously discovered) but through eth0 (thanks to MTS discovery)

Granted we could disable GE1 and GE2 in bios, but you get the idea. We'd like to be sure that even if discovery was made through eth0, eth2 gets the ip at the end of the provisionning process.

Feel free to say it I'm still not clear enough


--
Thomas HUMMEL


_______________________________________________
xCAT-user mailing list
xCAT-user@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/xcat-user

Reply via email to