(Corrected and Updated) Workaround for a Problem in Configuring Shared Ethernet Adapter Failover
sbodily 100000RUAA Visits (4996)
Until now, most documentation shows putting the Virtual I/O server's IP address on the SEA. The Virtual I/O server should have an IP address for convenience in reaching the server and to enable dynamic resource allocation.
In initial ATS testing, when we configured SEA failover following the new guidelines, we could no longer reach the clients. When we took the VIO servers IP address off the SEA and put it on a new, separate virtual Ethernet in the partition (not the one dedicated as a high-availability control channel, and not the server virtual Ethernet, but a new client-style virtual adapter), everything worked fine. In testing, the failovers were transparent and we didnt lose any connectivity by bringing down one VIO server or the other.
It turns out we must have misconfigured the first time, because subsequent attempts with the IP on the SEA have been successful.
But since using the virtual Ethernet worked, what about that option?
It actually could have some advantages. For one, the protocol stack used for the physical Ethernet and virtual Ethernet is different from the stack used by the SEA. In the future, a change to the "regular" Ethernet stack may or may not be made to the SEA. For another, it actually may simplify administration by separating the bridge from the device carrying the IP address. To quote a colleague in Austin: "One reason for using an extra virtual Ethernet adapter is for ease of management tasks. For example, if you want to change the MTU on the physical adapter used by the SEA, this can only be done if the adapter is not part of an SEA setup - i.e. you need to delete the SEA, change the MTU on the (now unused) physical adapter, and then recreate the SEA. If you were using the IP address on an extra virtual adapter, that's one less thing you'd have to dele