PAR4 - another data center metric? By peter judge

Views and Opinions on Green IT (May 23 2011)

1. 

   Just when you think you have met all the ways people are measuring data centre efficiency, along comes another one.

   PUE (power usage effectiveness), defined by the Green Grid (and nowinanewversion) pretty much rules the roost in terms of measuring the overall energy efficiency of a data center, But there have been plenty of discussions on how to measure and use it.

   Some of the effort to improve PUE looks upstream -how green is the electricity provided to the data center? The Green Grid’s CUE and WUE measure look into that.

   The harder issue is to look downstream, at what the electricity is used for within the IT equipment in the data centre. How much actual work do you get for each Joule of energy? This is a very significant thing to do, as it’s been argued that PUE is a blunt instrument, in danger of creating “false negatives” and unintended consequences.

   For instance, turning down the air conditioning will make servers run hotter and decrease PUE, which is good. But can also result in the servers’ built in fans working harder to cool the servers. The server fans are included in the IT power, so this makes PUE go down even further artificually - even though server fans are less efficiency than data center cooling systems. 

   Virtualising and consolidating servers can increase the utilisation of hardware - the proportion of time it is active. This might decrease the amount of energy going to the hardware, and actually put PUE up.

   Last year, I saw some interest around HUE, or hardware utilisation efficiency, an effort to measure how much processing you get done for a given unit of energy. This year, we have seen the launch of PAR4, a measure from infrastructure management company PowerAssure that attacks the same areas. 

   “Neither Green Grid nor EnergyStar have a standard that factors  hardware efficiency improvements into the rating and gives an annualized rating,” Clemens Pfeiffer, founder and CTO of Power Assure told me in an email.

   He’s not impressed with PUE which, he says, “favors inefficient servers as that increases the IT power  consumption. While that was not the intent of PUE it’s a reality that  more efficient servers usually increase the PUE.”

   I think that is an oversimplification, as inefficient servers need more cooling, so they can increase the overall power consumption as well as the IT power. Be that as it may, PAR4 has a formula:

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

   It is based on measurements of the power used when the system is running, and the idea is for manufacturers to quote it on servers as a sales tool.

   As such, it compares with SPECpower, which measures server performance in operations per Watt, and has been part of the EMA Energy Star program for some time.

   “It is a similar concept, however, SPECPower is using a Java implementation and lets each vendor run it on their own java virtual machine which makes results non-comparable,” says Pfeiffer (well, I think it is fine to let vendors pick their Java machine, as long as you want to compare Java performance).

   “Furthermore there is no rating in SPECPower nor an annualized adjustment based on Moore's law,” sayts Pfeiffer. PAR4 grades servers from green  to black (worst), according to defined bands which will go up by 50 PAR4 points every year, to keep pace with the expected server performance improvements.

   Power Assure says this adjustment is necessary, but I haven’t yet understood why. I would have thought that buyers would want to compare servers according to their actual power usage. Last year’s green-band server may have the same rating as this year’s but it will actually be less efficient, so this will make older servers look greener.

   Of course, it is true that older servers are greener than newer ones in one sense. Replacing them increases your carbon footprint when you take into account the embodied energy, but I’m not sure that is what Power Assure is getting at here.

   PAR4 has been picked up by analysts McKinsey, which has made it part of its CADE (Corporate Average Datacenter Efficiency) measure, which includes the overall utilisation of the hardware, and adjusts the IT power in the datacenter by excluding all servers which are idle - a simple but very effective way to reward efficiency and penalise idle servers which would otherwise actually decrease PUE.

   It’s also been picked up Underwriters Labs (UL), a body which certifies electrical equipment, normally for safety.

   UL will provide PAR4 measurements - under its own name, UK2604 - for IT equipment.

   “It is relatively easy to efficiently incorporate traditional energy efficiency testing as well as PAR4 into our normal safety certification related testing programs,”  Joe Hirschmugl, media relations manager for UL told me. “This provides our customer with a cost effective approach to getting the necessary certifications and/or meeting other regulatory requirements that may be necessary for their products. A better PAR4 rating will mean that products would be more attractive from an overall data center performance perspective.”

   However, this adoption should not blind us to the fact that PAR4 is proprietary. According to Hirschmugl, it 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.”

   At this stage, PUE is more general, but PAR4 going to be an interesting metric to watch. 

   Bookmark or Share this article

Recent Comments

Login to comment.