Virtualisation not as energy saving as it claims

I can't say that I want my servers to be unable to serve data for a period of the day. The problem is not letting the server sleep, it's ensuring it wakes up when needed. Wake-on-LAN doesn't wake up on seeing any packet, only on a special Magic Packet. Timed wake-up would be possible if you decided that your service would only be available between particular hours.

The hypervisor itself doesn't consume much power. It only runs when there is something for it to do: in response to clock interrupts, to reschedule which VM's virtual processor will run for the next time slice, and in response to a VM attempting to perform a privileged operation (the guest OS doing something that would require top-level kernel privileges, like writing to or reading from hardware). This might be as simple as redirecting a write or as complicated as emulating graphics commands, but even the most complicated are pretty simple.

One thing you should always be sure to do is install the hypervisor's support drivers and utilities in the VM. This ensures that a request from the guest to idle its virtual CPU is interpreted as stopping that VM from being scheduled. VMware calls this 'VMware Tools', Virtual Server 'Virtual Machine Additions' and Hyper-V 'Integration Components'. In addition these include device drivers which are a better match to emulated hardware than the standard drivers for the emulated devices, and in the case of Hyper-V, some devices are replaced with so-called synthetic devices which have no real-world equivalent, but map better onto the Windows API.

There really is a ton of energy saving in Virtualisation to be completely honest, from having just 1 powersupply (high wattage has higher efficiency), to having less hard drives (2 hard drives for the entire system vs 2 hard drives for each server), less space to cool... 

When processors idle they take up power while doing nothing, it's better to have them constantly stretched than sitting drawing watts for nothing.

Personally, I don't think his arguement makes sense. If you have 5 servers running at 5%CPU utilization and migrate those to one server running at 50% CPU utilization there will be significant power savings. This is not in dispute...

The article fails to take into consideration the benefit of consolidating hardware inefficiencies such as power supplies, fans, etc. 

If you are using the same processor a cpu cycle per watt profile is the same in a virt and non-virt system. If you run one machine at >90% utilization using a virtualization scheme it is better than running three non-virt machines at 30% utilization because you are saving on the power supply energy waste (rectification, heat, etc) and the wattage used by the cooling hardware like fans, etc.

Who the hell thought that adding another layer of abstraction would make things more efficient? Probably the same people that think the way to make business more efficient is to add another layer of bureaucracy. 

Mr. Barnum's legacy continues to live.

Indeed, any virtual vendor who claims to be saving energy on the executing box IS dishing out codswallop. But they can claim that by dishing out cycles amongst several virtualized servers when there is idle time, then the number of physical boxes can go down, which does represent "some" energy savings, I suppose, particularly as most power supplies are relatively inefficient at lower loads, hence, the more you load a box, the more efficiently it uses the juice.

I am rolling multiple processes that were previousl handled by more than 4 PCs onto one 1U server. If that is not saving money, I am not sure what is. Also, there are MORE processes now running on the virtual server than what was migrated and those processes would have needed more PCs if we continued doing things the old way.

The savings aren't JUST from the CPU. They're from:
1) Not having to the extra power supplies, fans, and other peripherals.
2) Not having to spend the extra money on hardware.

You can easily fit 20 vms on a DL380G5 which depending on your config would draw about 450 watts at about 50% cpu utilization. That is 22.5 watts per vm. Show me a physical server that runs at 22.5 watts per vm. The power and cooling savings is in consolidation of existing physical servers and stopping server sprawl.

... we discover that car pooling doesn't work because the car has to work harder to schlep 4 people than 1 person.


VMware has Distributed Power Management which can dynamically put hosts into suspend when fewer resources are needed.

Virtualisation saves a lot of power. You have failed to calculate the power draw of a server on idle. A server on idle will use about 60-70% of the power at full load.

So some rudimetary calculations show 100 servers with 500 watt power supplies at 70% utilisation gives 35000 watts

10 servers with 500 watt power supplies at 100% utilisation gives 5000 watts. 

Try downloading the HP power calculator and running some figures through it, you will quickly see that the power savings from increased utilisation are massive.

Having worked on virtualisation projects at multiple companies, I can say the virtualisation DOES reduce power consumption (not to mention space).

One specific example had a company who had completely used all of the power resources in the room. They tried to install 2 new servers and began to trip the power switches.

We installed 5 hosts, migrated 80% of their servers onto them (about 40 servers) and now they have freed up space and massive amounts of power. Not only that, the air conditioning that used to struggle to keep up with the amount of heat is using a lot less power.

I have installed a brand new dual socket quad core AMD server, that will be virtualising 9 older servers.

How is that less efficient?

Virtualisation is the best technology to port down to the PC/Server in a long time. Not only does it save tremendous amounts of energy as stated well in other comments to this article, it saves time setting up new servers, doing backups, and minimizing compatibility issues with multiple apps running on the same machine that require expensive IT resources to fix. You obviously do not know what you are talking about.

Guess that's why I don't read "Network World." Now if I can get theinquirer.net to stop spreading retarded information I'd be happy.

Recently I've just started researching virtualization therefore I believed above article 100%. 
But everyone's comments helped me to broaden my thinking. Thanks everyone for your explanation~

To Jeffrey Nonken, car pooling doesn't save fuel?

The weight of a car is greater than 1000Kg, and an average passenger will be less than 100Kg. Four people in one car means the car has to burn fuel for 1400Kg of weight. Four people in separate cars means all four cars have to burn fuel to move 4400Kg. looks like some flawed thinking on your part!