The case for NOT virtualising desktop machines.

I know a number of customers who have either virtualised, or are in the process of virtualising their desktop machines.

Okay. I have a number of personal opinions on this - as you can tell from the subject. They're mine and mine alone. I've plucked figures out of the air (Hey, I did used to be an IT manager you know) and the more observant will see that I'm not charging for power for the desktop PC in the office. Because thats a different budget, right? And I'm not adding in software, as you have to do that to both sides. And I'm not adding in the network cablling infrastructure cost - as you still need cables in the office as well as the data center.

Firstly, how much does a corporate PC cost. £400? £800 ? For 2-5 year lifespan ? Lets add £40 a month for the cost of someone to support this (a figure plucked out of the air). So over say 3 years, that'll be say £2,240 (assuming a really expensive machine - one that can play Unreal Tournament 20xx). Its a PITA as its out on the floor, on unprotected non-UPS power, and its hard drives will need to be backed up sometimes. But hey. Thats what it is.

Lets compare with a virtualised desktop. Firstly, lets assume you can get 10 - dammit, no 20 - virtual desktops out of a £5,000 machine. And each of those will need a SAN disk - and those aint cheap. Or indeed fast. Expect 'negative feedback' about the performance. Lets double the price for the SAN allocation and the virtualisation software cost (and associated experts). £10k.

And it'll sit in a data center, requiring UPS and dual power supplies. It wont be a low-power machine either. Electricity - instead of the 11p per unit (current-ish UK domestic price) at the desk, will be something like 40p a unit. (Those generators in the car park and the air-conditioning aint cheap, right?) And this machine will burn a *lot* of units. Optimistically 1kw. So thats 40p per hour - or £10,512 over three years. So now were' talking about £20k (give or take) for our monster machine for three years.

Of course this machine is spread over 20 people (low impact users - or 'folks who cant scream') - so thats a cost of £1k per user. And remember, those desktop support costs - say £40 a month - taking our per-user cost £2,440.

Hey. A whole army of bean-counters just creamed themselves. Nice one! Near cost equality! And its nice and centralised, so you dont need desktop support folks running around all these buildings. I'm sure that'll get some CFO a nice fat bonus.

Oh. Dear. Guess what. The 20 users a machine was a bit optimistic. We just added a really heavy client application such as an Oracle based database. Lets crank it back to 10 users a machine. So that'd be £2k per user for the machine, and our £40 a user/month for support. Ooops. Thats now: £3,440.

But hey, look - the environment is locked down. Everything is centralised. Isnt that cool? And the virtualised environment means that we can just click and drag machines around! (Well, thats not the case in reality).

Oh dear. The machine performance is appalling. Only 5 power users a machine now. What does that do to our figures?

£4k + £40/user/machine = £5,440. Woops. Thats hugely expensive. Like double-ish.

See how easy that was to do ? And what do the users see ? A huge amount of inflexibilty. Huge inertia. Peak periods where machines grind to a halt. A huge battery of servers in an expensive data center someplace. 

Imagine if it were so bad that you'd need a dedicated machine per user ? £21,440. 10x?

(I've missed out: Additional network costs from each office to the data center, and the cost of thin clients on each desktop. Monitors + keyboards + mice would be used in either case. Call that another £200 or so in capital expenses per user?)

Right. So yeah. Run your figures through a process like this and be aware that this can very quickly become a huge political millstone around the sponsors neck. Ouch. Remember - for this to be economically viable, everyone has the same build, the same performance, the same disk space. And this has to run for three years. 

Risky. You'd need really big crystal balls to pull this one off.