What Is Behind This New Efficiency Measure?

Last week, we heard about a new measure of efficiency, which claims to rate the efficiency of servers according to the number of transactions they provide per Joule of energy supplied. I think it raises serious questions.

The PAR4 measure, devised by infrastructure management company Power Assure, is supposed to help data centre owners choose what server to buy, as they will come supplied with a PAR4 score, measured by Underwriters’ Labs, a certification body which normally concerns itself with electrical safety.

Now, this could be interesting, because the Green Grid has said it wants to help measure data centre efficiency with more than just the broad brush approach of PUE (Power Usage Effectiveness). PUE measures how good your data centre is at delivering power to the servers, instead of to equipment which cools them down.

If people really want to use less energy for IT, then they will need to look at how much actual processing gets done by the energy that reaches the racks – and PAR4 appears to be doing just that.

Not many details yet

But so far, there is not much known about it, and given that we don’t approve of duplications in green standards, we have doubts over the fact that it seems to overlap with the  SPECpower measure from the Standard Performance Evaluation Corporation (SPEC), which has been measuring server performance in standardised “ssj_ops” transactions per Watt since 2007.  SPECpower is part of the Energy Star programme for servers, which has backing from the US government’s environmental protection agency (EPA).

Power Assure answered some of my questions, and sent me a white paper on the relationship between this measure and other green metrics, but left me with more unanswered issues.

Firstly, the white paper, called “Combining PUE with other energy efficiency metrics”  is not online anywhere I can find. It includes an equation for the PAR4 measure, which is as follows

PAR4=log.sub.2(tps/watt)*100

This just details how to take a measure of transactions per second per Watt  (tps/watt) and map it to a logarithmic scale, but it doesn’t say where the tps/watt firgure comes from. What benchmark is Power Assure using?

SPECpower uses a Java benchmark running on each vendor’s own Java machine. Clemens Pfeiffer, CTO of Power Assure has criticised that, saying it make things non-comparable – but his objections would apply to any benchmark and presumably also to PAR4’s.

Secondly, PAR4 is adjusted so its quality bands go up by 50 points per year, to compensate for increases in server power driven by Moore’s Law. This makes very little sense to me. I would have thought users want to compare server power, without adjusting that to determine how well the server guys are doing at getting faster.

If last year’s server had an easier run on PAR4, the adjustment effectively weights the green scale in favour of older servers, which makes little sense.

Thirdly, Power Assure overstates the case against PUE, claiming it rewards inefficient IT, claiming that “more efficient servers usually increase the PUE”, something which isn’t really true because there is a linkage between power used by IT and the cooling power required.

A proprietary scheme?

Strangely, Power Assure refers to “hot isles” instead of “hot aisles” in its paper, but I am not on good ground to criticise other people’s typos.

Despite all this, Power Assure claims that PAR4 is already involved in McKinsey’s CADE (Corporate Average of Datacenter Efficiency) measure, something I’ve not been able to follow up in detail yet, and it has the UL endorsement we mentioned at the top.

The biggest issue, however, is one that UL’s media relations manager Joe Hirschmugl mentioned to me almost in passing. PAR4, said Hirschmugl, is “directly related to the the data center management scheme that Power Assure provides to its customers.  The PAR4 rating and associated data specifically supports the parameters used by the Power Assure scheme.”

In other words, to use PAR4, you have to be running the Power Assure products.

PAR4, then, appears to be a proprietary scheme, a sales tool, which duplicates and appears to misunderstand existing metrics.

I don’t believe measurements should be exclusive to any one technology. We don’t have to buy rulers from one source. So, until we hear more, I think we should all be wary of PAR4.