i have found the reply alone, but tell me if it is correct for you:
dual-vios setup with SEA failover, NPIV and sdmc, create dlpar with 2 fc virtual adapter 1 in vios1 and 1 in vios 2, create 2 virtual ethernet adapter with priority order following the primary/standby sea.
then i suppose that if i shutdown 1 server(and 1 vios) nothing happened to dlpar because it can access disks and network.
or i'm wrong? i mean i need to migrate 1st the dlpar to another physical server to failover?
this is the point that i cannot understand well because i didn't have direct experience.
Forget the DLPAR as this is not necessary on a new configuration and can cloud the issue, consider this example:
You have a brand new Power 7 frame on this you need to build 2 VIO servers and a client LPAR. The secret of an HA design is resilience and with the dual VIO configuration this is what you will get.
It is still a "best practice" to use internal disks to boot the VIO servers but nothing stops you using SAN provided disks. As you are using blades the internal disk configuration has a flaw as there are only 2 internal disks (this would be 1 disk per VIO with no mirror possible) per PS700 blade server, also the disks are NOT hot swappable.
For this eample we will build the LPARs on a P720 and therefore we will use a pair of SAS disks for each VIO server for the O/S boot device using a normal AIX mirrored pair. All client disks (O/S and data) will be SAN attached using NPIV created virtual fibre channnel adapters. Also all networks will be supplied by virtual Shared Ethernet Adapters (SEA) provided by each VIO server.
Once the VIO servers have been built the virtual server fibre channel adapters are created and the necessary virtual SEA adapters are also created (remembering to set the Trunk Priority on each VIO server for correct traffic flow and failover). This activity of course takes place on the HMC. I actually notice that you are proposing to use the IBM PS700 (PS702 / PS704) blades, these can only be controlled by the new Systems Director Management Console (SDMC). However the creation of the virtual devices is much the same as with the HMC. One thing I forgot to say above is any Etherchannel adapters should be created prior to the creation of the virtual SEA adapters so as to keep the ENT numbering sequential.
The client LPAR profile can now be created as both the VIO servers are running. Create the four virtual client fibre channel adapters (2 per VIO server O/S and data). Then create the virtual client Ethernet adapters. Once the virtual WWNs have been zoned into the SAN storage and the LPAR has been setup in SMS to boot from either the NIM server or DVD, the client LPAR can be built.
When the O/S has been installed a simple
lspath
should show the storage connected by 4 paths as follows:
This is a fully resilient LPAR which is immune from a single path of failure due to the dual VIO construction.
What benefits would HACMP hold for you?
This is debatable as you are using blades! If both your blades were fitted into the same blade centre then would there be any point in an HA? Fail over would be to the second blade. As both blade servers would be constructed in a dual VIO server there are only a handful of reasons why the environment would fail; loss of SAN etc. This would mean the only real possible failure would be a complete loss of the blade centre which, of course would affect both blade servers!
VIO and HA are completely different concepts. With VIO server hardware resources are shared. If you have single Blade server and you want to have multiple client partitions then you can have VIO. Dual VIO can eliminate the downtime for hardware failure,as you client partition can be accessible from both vio.
Now if you have complete server hardware failure for i.e. servers planner goes faulty then your both the vio will be down with you client partition also. In this scenario if you have HACMP configured then your database will be up from another hardware. For HA you need tow physical servers or lpars from different physical servers or CEC's. Then you have to configure HACMP so that if one server is down your database with your ip address will be moved to another node configured in HACMP.
thanks VJM for the clarification, infact i will use powervm for the virtualized enviorment with dual vios for hw redundancy and for my client partition i will be use oracle rac cluster on aix and linux to manage the high availability without hacmp(powerha). all those things managed with sdmc on a vmware cluster.
ibm has certified this solution for their product and oracle has certified Rac for HA on Powervm and aix.
if you have another advice or something else pls let me know.