The whitepapers are good regarding redundant VIO. However you need to evaluate what Service Level you need to reach in order to decide whether tha additional management overhead of the second VIO Server is worthwhile.
In the case you are talking about here you'd need to have a machine with at least 2 SCSI adapter in it as you need to allocate a SCSI adapter to each VIO Server. Then each VIO Server needs disk space for its operating system. After that any disk room you have left on disks allocated to the VIO Server can be connected through to the LPARS. You can do this using full disks or logical volumes on the disk.
So really you need a mimimum of 2 SCSI adapters and 6 physical disks before you can implement this. In my ivory towered world I don't want to put Logical Volumes of the root disk of the VIO Servers for other Operating Systems. By perferance though I want to boot LPARS that use the Virtual I/O Server from SAN disks and not local disks. In this case VIO is at its strongest (in my not very humble opinion). The white papers only use this method.
If you do want to use SCSI volumes from each VIO you are left having to repair mirror problems frequently if you reboot a VIO (after an upgrade for instance), whereas using the same SAN disk allocated through both VIO servers that problem is not there.
What isn't discussed is the spagetti that comes from allocating your Virtual I/O slots. This is one place where you need to have a procedure in place to manage these manually. You allocate them so you need to be structured, otherwise its a mess, and virtually unmanageable. Do not be tempted to simply allocated them in numerical order.
If it helps I use: under 100 for Ethernet, 100-199 for Virtual SCSI from VIO1 and 200-299 for Virtual SCSI from VIO2. We then allocate slot 100 and slot 200 to the same LPAR - and the LPAR relate those its own slots 100 and 200.
The Ethernet slots - I always make the VLAN id equal to the slot if, ie slot 20 will always have VLAN 20 attached. It saves a lot of debugging headaches...
Someone asked me recently - if all the LPARs are booting from SAN why not the VIO itself? My answer is thatif there is a big SAN problem I can then still start the VIO and debug the SAN from there - If I cannot boot the LPAR I have nothing.
__________________
Ross Mather, IBM AIX IT Specialist.
That said anything I say here is my own opinion and not anything that you can ever hold against IBM.
Ohhh and don't forget that I make mistakes too.... |