Computron, please tell me if i am trying to explain to you something you do already know, but i sense the problem is in fact the understanding of how virtualisation the IBM way works.

We'll start at the basics: what you get from IBM is a "managed system", which is the synonym for "hardware node" in most Linux-centric virtualisation platforms (Openbox, OpenVM, KVM, OpenVZ, etc.). The difference between these Linux-systems and the IBM hardware is that there is no OS you have to install at the base. Everything is handled in (or near) hardware.

With i.e. OpenBox you install an OS to the system, install the virtualisation software onto it, create several virtual machine profiles and start these as programs in the base system.

With IBM you have the hardware which already allows virtualisation (controlled by the HMC, which is a mere management platform, not a direct part of the virtualized system) and install directly into the LPARs (=virtual machines) you create.

For this to work you need to create "profiles", where you give out parts of the systems resources to the various LPARs. Your system might come with 2 processors and you might decide to give 0.1 to the one system and 1.2 to the other, keeping the remaining 0.7 in reserve. So, a good idea would be to first meticulously describe your system here:

# of Processors
# amount of memory installed
# and types of network adapters
# and types of built-in disks
etc.

This will get you a base for planning what to give to whom and when.

Now, the virtualisation works well for "anonymus" ressources like CPU and memory. It gets a little more tricky with less anonymous resources, namely network adapters and (internal) disks. Therefore the first (usually) thing to do is to create a special LPAR which acts as a man-in-the-middle between the physical hardware and the (other) virtualised systems. This special type of LPAR is called "VIOS" (virtual I/O server) and one has usually 2 (or several pairs) of them (a stripped-down two-node HACMP cluster is part of the software package for VIOS) per managed system.

Here is a typical way to create a system:

after discovering the MS and getting the HMC to manage it you create two LPAR profiles for your VIOS and use "installios" to install the VIOS software onto it. Typically these are the only systems in the whole MS which have physical disks to boot from. In a typical setting the bootdisks for all the other LPARs come from SAN and are connected to the LPARs via the VIOS as vSCSI LUNs.

You also give the VIOS each one (or more) real network adapter(s) to connect to the outside. Network traffic for the other LPARs is done via the VIOSes which act as a (virtual) bridge. You will have to plan thoroughly how to route what through this. Non-VIOS LPARs usually get one virtual network device which is in turn created on the VIOS and mapped to the system. (This is quite similar to how the Linux systems handle things, except that there is no specialised virtual system there but the hardware node, where you define this.)

The VIOS can pass information about handed-down LUNs and virtualised network to another VIOS - even one in another managed system - and it is possible to move (running!) LPARs around between different MSes. This is called "Live Partition Mobility" and a thing not possible with any other virtualisation system.

Also the built-in hypervisor is the fastest in the industry. It offers a real huge bandwidth and in practice you will have nothing to do to tune it.

So i suggest you start by posting your exact system data and start to plan what you want to do and how.

I hope this helps.

bakunin