The analysis in the article is valid only if the server OS is running using just ACPI and standard power management. If CPU frequency scaling is being used math is different.

Virtualisation frameworks are not very good in supporting CPU frequency scaling which is the biggest energy saver in a datacenter. If you virtualise you usually have to run the host OS at full CPU frequency. If you do not you can run it at on-demand scaling (if supported by OS). This for a Xeon setup may mean anything of up to 200-300W per 1U server or up to 100W per average blade.

As a result if you run under 5-8 VMs per server (which is the usual number), virtualisation does not really save a lot of energy compared to a well tuned correctly run datacenter. Any savings are mostly from "less iron", but not from "less energy".

Funny, because in Finland, there's an extremely bad word ('vittu') which is the equivalent of 'fuck' in English, and where English-speakers might refer to "the F-word", Finnish speakers would refer to "the V-word". So for a moment I wondered what this article was about...

But back on topic. Surely if you outfitted those same virtualised servers with 'green' components (energy-saving PSUs and fans, low power consumption hard drives, mobile prosessors, and so forth) then the comparison of four servers at 20% capacity vs one server at 80% doesn't really pan out. If those four servers had green components from the start, it might be different, but a single, green, virtualised server surely must be better than the alternative.

"Does virtualisation = energy savings?"

No.

Virtualisation = Energy Generation.

I trust that is not ambiguous? QuITe how IT is done, is of course, on a strictly Need 42 Know Basis between Trusted and Trusting Computer Circles. Circles/Networks which I suppose you can Imagine are ensuring that IT is appropriately Monetised and Secured.

More anon as IT Progresses. Good Day.

The empty space is usually a saving, it gets utilised for something else rather than sitting idle, which saves the company even more money as they don't need to build more datacenters (which means the World also stays greener). In London I hear floor prices of £50 to £150 per square foot for office space (I'm not an estate agent so don't email me if those prices are low!), that's before you add the cost of converting it into a datacenter room.

Then you have the green savings on associated admin costs - virtualised resources are also usually centrally managed via software, reducing the number of car journeys out at 3am to reboot servers that some numptie has crashed, because now you can sit at home in your pyjamas and do it with a few mouse-clicks (that's if you haven't automated the whole thing anyway). BOFHs reaching for your cattleprods, the secret is out, you will lose a lot of that cushy quad-time call-out money!

Possibly, but the article also acknowledges that one of the largest power consumers in the Datacentre is air-conditioning - surely, if simplisticly, fewer servers -> less space required -> smaller datacentre -> reduced air-conditioning footprint?

"It seems slightly ironic that a software solution (virtualisation) may play the largest part in solving what is essentially a hardware problem."

reminds me of the old lightbulb joke:

How many hardware engineers does it take to change a light bulb?

None. "We'll fix it in software."

The BCS Data Centre Specialist Group have been investigating this issue in some depth. The load to power relationship of servers varies greatly, but none are linear. There are many more components than just the processors that vary their power with system load and programs already underway to expand this. We are, however, not likely to see linear either, due to the 'power floor' of components required for the server to be running at all.

There is significant benefit to be had from virtualisation but this, again varies. High overhead virtualisation that invokes a full copy of the operating system for every application, whilst an improvement, is fundamentally limited and does not represent a signficant change in mind set from 'one app per server'. Low overhead virtualisation such as Solaris 10 containers can go much further and provide much more benefit.

The impact on the data centre infrastructure and power bill is also rather more complex, whilst traditional cooling systems will not react much, the power delivery infrastructure will and the impact will be visible on the bill. More modern data centre design approaches will react very differently to this change in load density and profile.

The real savings do come at the data centre level. The hardware cost of the server has been substantially exceeded by the cost of the data centre power infrastructure to support it for several years now, with the power cost expected to exceed the server cost within a couple of years. The price of a server is increasingly irrelevant in the cost of ownership equation.

1. There is a significant reduction of metal & plastic in a virtualised environment in comparison to a non-virtualised environment.

2. The virtualised environment requires less human labour, transport costs and fuel usage to set up. The same applies to upgrades (which are no virtual too) and future server deployments.

3. The ability to migrate a virtual server to another country without any physical movement of parts.

In short, less metal, less fuel, and easier management. Also, while I agree that it's unwise to compare a good VM infrastructure with a badly set up datacentre, it's still pretty obvious that a server room that has been designed to cool you virtualisation kit is going to be 'greener' than a larger server room with a larger number of physical servers in it.

That's the beauty of "green" reductions, it's nearly impossible to prove or disprove - there are knock on effects all the way down the line.

For example, have we considered:

* Running these machines into the ground as opposed to dumping them before their end of life? There are "green" implications for the recycling effort, the delivery and manufacture of new machines/components

* Going forward, less machines to be decommissioned/upgraded

* Uber machines will need to be upgraded more frequently as they'll have less room for exansion

* Running a machine at 80% load continuously will reduce the MTBF figures

I'm sure anyone could come up with a convincing argument either way. Brings me back to my first point, that's the beauty of "green" compliance.

I would have to say the article was severely flawed by the "numbers" being thrown out.

What happens with the old and unusable parts? Will they be recycled?

Probably not. Crushed up, melted down to slag to be used as landfill for the next roadway or building. Not really green now is it?

...Live In Space

I don't believe any part of this article because its all based on overly simplistic theory. What we need is measurement:

- leave the recycling and space saving issues to one side and just measure power consumption

- set up 4 servers, each running a different workload, e.g. web server, database, office file server and SAP office admin.

- adjust simulated workloads so all are 20% utilised

- measure the power consumption for the complete "data centre"

- put the lot onto one of the servers, one per VM.

- turn the other servers off.

- check that the virtualised server has at least the same throughput as the original setup.

- measure power consumption for the complete "data centre" again, using the original workload.

Now look at the results. If the virtualised server can't handle the original workload, the virtualisation FAILS. If that hurdle is passed, report the power saving.

Any reputable virtualisation supplier should be able to do this comparison quickly and cheaply.

I for one will disbelieve any and all power saving claims until such a test has been done and the results reported.

As someone pointed out, it is very possible to lower AC requirements - we've done so ourselves, reprogramming the AC unit in our server room to less than half the previous output.

The other factor with AC is quite simply less machines = less heat = less power required by AC. Yes the one VM does require more power than a single server, so it's not (in our case) 1/4 of the heat output, it was more like 3/4.. but that was still a significant drop.

Generating 25% less heat combined with reducing the output of the AC to about 1/2 the previous room coverage = a nice saving on the electricity bill. I'm only guessing here, but I reckon a lower electricity bill is quite a good indicator that our setup is greenerer that it was before.

*greenerer is a technical term I'm lead to believe, meaning more trees for George Bush to stand next to in order to look like an environmentally friendly prez.

I work at a company which builds electronic devices and I find it plausible that the production of a computer takes way more energy than it's operation.

Just look at a typical motherboard. Every little part is packaged in a lot of plastic which gets thrown away. The soldering process takes quite a lot of energy and many boards need to through it several times.

So what you need to minimize is the amount of new hardware buildt.

BTW, power supplies tend to be more efficient when evenly loaded. If you can guarante a certain amount of load, you can even use some with a far higher efficiency than typical computer PSUs.

I think there has been some great input in all the above but have to say the article/heading/topic does seem a little misguided.

It does seem a bit like asking if a paintbrush is artistic or not! ... i.e. virtualisation is an fantastic enabler for change, this doesn't mean it is green in itself, nor does it mean anyone who uses it will automatically gain environmental impact advantage.

I do wonder if this was deliberate just to park the reactions?

...makes me laugh.

They say they've cut the carbon footprint at their Ellesmere Port plant by 30%

Could this be because they're making 30% less stuff there these days?

http://news.bbc.co.uk/1/hi/wales/north_east/4994814.stm

As someone famous once said, there's lies, damn lies and green statistics...