Data centers could get 60% better performance from their infrastructure if they ditched their hypervisors and ran bare metal, according to a new study, which claims operators can get a lot more processing power per pound if they scaled down.

The study came in the form of a benchmarking exercise by infrastructure specialist Bigstep.

Among its more controversial findings it claims that not only are upgrades unnecessary in order to get more throughput, but performance can actually be increased by removing layers of software and hardware.

According to the study most data centers are not getting the full benefits from the equipment they have invested in.

The studies assessed areas of potential performance improvement in infrastructure, using Linpack, SysBench and TPC-DS standard hardware benchmarking tools.

The benchmarking investigated the effects of four actions: deactivating central processing units (CPUs), the choice of operating system (OS), improving memory frequency and hyper-threading.

The performance boost created by each action was measured as 15%, 20%, 20% and 10% respectively.

The Bigstep tests found that single processor bare metal instances can generate better performance than dual processor machines, all other things being equal.

So adding a second processor to a machine can lower a database’s performance and conversely subtracting one can boost performance.

This limitation can be overcome by configuring the OS to use only local memory for each processor.

Meanwhile, the choice of OS makes up to 20% difference in performance, owing to the differences between kernel versions and settings.

Similarly, a performance dividend can be enjoyed by improving memory frequency.

This is because memory is routinely assumed to be fast enough – when it often isn’t.

In Linpack benchmarks, Bigstep found that replacing 1333 MHz dual inline memory modules (DIMMs) with 1866 MHz DIMMs from the same provider increased overall computing performance (measured in GigaFLOPs) by 20%.

Finally hyper-threading is often counter-productive and decreases rather than increases performance, the study concluded.

The findings can be applied to any infrastructure said Bigstep commercial director Ioana Hreninciuc, who called on data centers to get more for their money.

“Many users are not seeing the performance they could and, with just a few tweaks, any server could perform by up to 60% better, avoiding costly upgrades or migrations,” Hreninciuc said.