It’s an age old ‘virtual’ discussion and one I’m working with at the moment. A raft of virtual machines all *needing* 4xCPUs but when challenged no qualification can be evidenced.

Stuck on the motorway this morning this analogy leapt to mind.

Consider, you’ve organised a road trip holiday, more than likely you’d plan like this.

Yes, good idea

1. Full tank of fuel
2. Pack a kit bag with your clothes and essentials
3. Pack enough food to get you started and keep you running for a predictable while
4. Subsequent stops during the tour would be for re-fuelling to keep the balance going
5. A happy medium would be found

That would seem a sensible approach and would be similar to adjusting the resources within a Virtualised Infrastructure to meet the demands of a virtual machine to be shaped around business needs.

No, bad idea

A somewhat daft approach would be:

1. Divide total mileage by the estimated range in the tank and realise you need 12 full tanks
2. Accommodate the 12 tanks by stripping out the seats out and filling the vehicle with jerry cans
3. Calculate roughly how much food you’d need for the few weeks journey
4. Pack boxes and boxes of food (some perishable)
5. Clothes, well they may well fit into the glove box…
6. Additional passengers? Hmm, squeeze in and sit on the lap of the driver or jerry cans

Top heavy virtual machines?

Throwing everything in one go really doesn’t work for a holiday so why apply the “I need all resources immediately” for a virtual machine requirement.

I’ll be using this analogy again… and again… and again…

Where do you start when designing Resource Pools in an empty VI?

a) Create pools based on Development, Test & Live then divide total resource by 3?
b) Divide total resource by a rough split of your planned P2V phasing?
c) Create pools for logical grouping of VMs?
d) Group by vCPU or RAM allocation?

etc… Anyway, I think I’ve painted the picture. So what’s the answer?

Simple, understand the business needs.

“But that’s not a technical answer!”

A Virtualised Infrastructure is really no different to any other service provision. If you’ve just moved home you wouldn’t ring up your local water authority and request ‘x’ amount of water must be provided to your home. You’d say “I’d like water please and let me know how much I owe you for a residential supply.”

Why not apply this methodology to service provision?

Example, you need a single SMTP relay running under IIS 6, anticipated 10k mail a day. POC this with a test tool, review utilisation demands. Snapshot results, add a bit more based upon forecasted peeks, done. The project now requests what it needs to deliver the requirements.

Resource pool decisions don’t necessarily have to be final. Shape the delivery of services around business demand. Force new projects to fully comprehend their requirements and impact.

We’re in a strong position using a Virtualised Infrastructure:
- Resource can be ramped up, down and constrained. Whereas fast executed projects using physical service provision can be disastrous if poorly executed.

Resource allocation with a sliding scale – it’s the way forward.