Storage virtualisation (IBM's SVC, EMC's InVista, HDS TagmaStore) needs to be ILM-aware in order to use multiple tiers of storage. What I mean, in a nutshell, is how does the virtualisation engine know on which tier to store an individual piece of data?

Let's suppose we have a virtualisation engine (VE) running and also ILM with three disk tiers: FC; SAS; SATA. A piece of data comes in to the VE. It's a tier two chunk of data. How does the VE know where to put it?

The application/server-facing VE represents the disk storage behind it as a single, logical pool of storage. So an application on the connected servers sends it a chunk of data to store on tier two. But the VE doesn't represent a multi-tiered storage pool to the applications. It represents a single pool.

Do the applications then know about the multiple tiers and send the data with a tag saying (tier two please)? Even if they did the VE has to know about tier two and has to divide its single logical pool into pools per tier.

How is this done? Can you set up a VE so that it knows about multiple tiers of storage? In that case, logically, it has to represent multiple logical pools of storage to applications. How do the apps detect this and use the information?

It's a nonsense for applications in servers to be tier-aware and to send read and write requests to the VE with tier tags attached.

So where does the tier decision get made? It can't be made on the server-facing side of the VE because the VE projects a single tier of storage. I don't believe it can presently be made downsream of the VE because there is no space. The VE talks directly to storage arrays.

So, let's try this, logically, it has to be made upstream of the VE with another storage function (ILM data placement engine) sitting between the applications and servers making data I/O requests and the VE. This data placement engine tells the VE where to place the data - only the VE projects a single logical of storage. Back to square one.

My conclusion from this is that ILM simply cannot work in a virtualised SAN storage environment unless there is virtualisation of each ILM tier and the ILM data placement engine (DPE) knows about this and directs I/O requests to the right - and virtualised - tier.

In effect this means, to my mind, that the DPE and VE have to be tightly integrated. You would define ILM tiers to the DPE and it then tells the VE what tiers have what drive arrays/disk types in them and please go virtualise them.

We just made virtualisation much more complicated and probably decreased disk utilisation efficiency as there will be different utilisation efficiency per tier.

At present then, it is not possible to have ILM in a virtualised SAN - or NAS - storage environment. Am I right or wrong?