Power Usage Effectiveness (PUE) has been a tremendously useful metric for the datacentre so far, but it is not a complete measure of data centre efficiency, according to BCS, the Chartered Institute for IT.
PUE compares the total energy consumed by a data centre to the amount of energy that actually reaches the IT equipment, showing how much is lost to other equipment such as cooling systems. In other words, measurement stops at the IT power cord.
While the metric successfully addresses the infrastructure part of the stack, it does not tell us anything about the IT equipment, the software running on that infrastructure, or the value of the business processes running on that software.
According to Gary Thornton, Technical Director at CNet Training and member of BCS, a lot of work is being done at the moment around using “IT work” as a proxy for useful work in the data centre. However, the problem with this, according to Thornton is that the two are not quite linear.
“Proposals that suggest using the IT draw of the equipment make the assumption that IT power – any IT power, irrespective of what it's doing – is useful. Ask yourselves, is that really the case?” said Thornton, speaking at The Green Grid EMEA Summit in Brussels.
“The energy associated with the actual useful IT work is around 40-60% when the server is idle and doing nothing. It's a reasonable proxy because it has the same characteristic – it rises with the IT work – but it's not accurate.”
Even with virtualisation, servers are running at around 40-50% on average, so the error is significant, and gets worse as server utilisation decreases. Moreover, using IT power as a useful work proxy rewards poor utilisation, because the IT power at poor utilisation is actually higher.
“We can accept those errors if we're using IT work as a proxy, but if we're trying to use it for accurate measurements then it's just not cutting the mustard right now,” said Thornton.
“The chaps within The Green Grid that are working on these new metrics have got significant challenges in getting to the bottom of this.
BCS has come up with its own metric to try and solve this problem. Instead of looking at IT efficiency, the Fixed to Variable Energy Ratio (FVER) metric focuses on inefficiency, and targets the wasted energy that's not doing useful work.
FVER defines useful work by calculating the ratio of fixed energy (the power draw that would exist even if the data centre was inactive) to variable energy (the amount of power needed on top of this to carry out useful work). The definition of “useful work” is determined by the operators themselves, based on productivity and business value.
“Business value is subjective, it's user-specific. Each business is different, so it's not comparable between businesses. Why should it be?” said Thornton.
“So what we propose is you allow each operator to select what is relevant to them. Each data centre might carry out several hundred different functions, but the business can select which are the big hitters that bring most value to that business.”
Once useful work has been defined, then the operator can monitor this over a week, and compare it with the overall power consumption for the week. Using the FVER formula, they will come up with a number between one and 10.
Thornton said that a FVER score of between 1 and 1.5 is ideal, between 1.5 and 2 is good, between 2 a and 4 is OK, and between 4 and 10 is poor.
“FVER and PUE actually complement each other very well, because FVER attacks the overheads where PUE doesn't reach,” he said. “If we've got high overheads there's got to be an opportunity there to reduce those overheads, and that has been the success of PUE.”
Thornton admitted that FVER is not an absolute measure of data centre efficiency because it does not calculate the number of useful instances and divide them by total capability of work – which is what some metrics do. But it does help break the data centre down into manageable chunks.
“A data centre is a complex ecosystem of technology, people and processes; so how can we tackle that? One bite at a time,” he concluded.
More information on BCS's FVER metric can be found here.