First, I work with some guys who have a bias for other unix flavors, mostly HP-UX and Sun. So, I'm taking a beating over this stuff right now, and any thoughts anyone can share on this would be greatly appreciated.

Here's the set up:

p570 (8 procs, 18GB RAM, 6 x 73GB hdisk).
2 VIO servers (3 x 73GB hdisks each).
5 non-VIO LPARs (all disks assigned by VIO).

We have some storage coming off an EMC for two of the LPARs and a possibility of more storage for other LPARs. There are 4 FC adapters in the machine, divided equally between the VIO servers.

The EMC storage is assigned to the VIO server, since it is the one that controls the FCs. The vio server hands out the storage to the appropriate LPAR, which sees the whole big chunk of storage as a single disk.

Each vio server can see the others SAN storage, so in the event of a failure things can be shifted over to the functioning vio server and everything can carry on as normal.

But the real question is: is this the best set up?

I've found a couple of redbooks that touch on the topic, and they elude to this set up or something similar. What I would really like to see is some sort of automated failure detection that would shift things around between the VIO servers in the event of a failure. HACMP for VIO, if you will.

I hope someone out there has thought about this as much as I have. Until I can get a good answer for this I honestly cannot recommend this go into a production environment.

I guess that three lpar's will access the disks through the first vio, and the two other lpar's will access the disks through the second vio.
It's probably the best way to share four FC adapters between five LPAR's.
Personnally, I would use ten FC adapters, two of them directly owned by each LPAR, and I would use the virtual IO server only for the system disks and Ethernet adapters.
This would be the really efficient way. Unfortunately this would cost ten FC adapters instead of four. But it would be really safe in a production environment.
The EMC data security would be made by PowerPath.
The system disk security would be provided by AIX mirroring between the two virtual disks, one on each virtual I/O server.
The Ethernet security would be provided by Etherchannel (the main adapter on the first IO server, the backup adapter served by the other I/O server), be sure to plug the two cables on two different switches.
Fully operationnal, no need of an extra HA/software, only long-ago prooved mechanisms.

The only thing out of your suggestion that I don't have right now is the ten FC adapters. Really I would only need eight. Three out of the seven LPARs are "utility" servers. Two are the VIO servers and one is a NIM server. The remaining four will need the SAN storage.

But thanks for the thoughts. That is pretty much what we came up with.

Four LPAR's, four FC adapters, and you can't have one adapter per LPAR if you don't want any SPOF...
By the way, one adapter per LPAR and HA between partition in case of DISKERROR3 ? this should work !