Organizations with many servers are increasingly interested to know just how much energy they could save by upgrading to new hardware, but it is often difficult to find solid data about this kind of energy consumption. Now, detailed tests in The Hypervisor Labs have quantified the potential savings.
Our test results show servers fitted with new energy efficient CPUs consume about half the electricity when they are idle compared to older counterparts.
Also, new servers run software more quickly, so they spend less time running at full speed compared to older kit. Our test results show running at full speed consumes about double the power of being idle, so it’s best avoided.
Finally, the above energy savings mean newer servers produce less heat, which in turn means air conditioning equipment needs to do less work to keep datacenters cool.
How we tested
We used three servers with two CPU sockets, and similar RAM and hard disks configurations. We installed the same software on each server. The main difference between the systems was in the type of CPUs each used.
The first server was a Dell PowerEdge 2850, which was built about five years ago using single core 3.6GHz Intel Xeon processors. The second server was a Dell PowerEdge 2950, built a few years later using quad core 3.16GHz Intel Xeon X5460 processors. The third of our servers was a modern system fitted with energy efficient 2.93GHz quad core Intel Xeon X5570 processors.
We installed the same version of Microsoft SQL Server on each piece of hardware, and used a SQL load generator to run the same set of complex queries against each server. We configured probes on the mains power supply to record the each server’s power consumption during the tests. The results are here.
Our test results show clearly that modern servers use around half the power of equivalent older hardware while the CPUs and RAM are idle. The power consumption of busy hardware is broadly comparable, although each generation of server hardware we tested used approximately 10 percent less power when busy compared to the previous one.
Another significant energy saving comes from the fact that modern CPUs run software more quickly than older chips. The oldest equipment took about eight times longer to run the same software workloads compared to the newest server we tested.
Any financial savings from the use of energy efficient hardware would obviously depend on the price paid for electricity and the number of servers in use.
In the UK electricity is sold in units, each being equivalent to 1 kilowatt hour (kwh). The oldest servers in our tests used 230watts when idle, or 0.23 kwh, or 2,015kwh per year.
If you pay 10p per unit, switching off that device would save £201.50 per year. Clearly, if you were able to consolidate the workloads from 15 servers onto a single new server, you’d pay around £3,000 per year less in energy bills – a big enough saving to buy a new server. This calculation does not take into account the savings that would probably accrue from reduced air conditioning costs.
Server virtualization tools could be used to make this consolidation project quick and easy to complete.
Older servers that regularly run at 100% CPU utilisation should obviously also be upgraded. Many such systems would be suitable for virtualization, and migration from old hardware to new virtual infrastructure could be done automatically using physical to virtual (P2V) migration tools.