There was a good turn out for the multiple JISC funded project workshop on Metering and Managing Energy Consumption in Data Centres held on the 12th July in Leeds. Delegates from a good number of universities across the UK, and a visitor from Dublin came together to discuss the issues around what to meter, how to meter it and how to use this data to drive improvements in data centre efficiency.
This workshop was a joint venture between these JISC funded projects:
First up was Roland Cross from Leeds Met, who described their work to improve their currently inefficient data-centre, while taking the time to meter and measure things to see just where the power is going.. They are tacking this through a combination of power upgrades, changes to air handling, hot/cold aisle separation and further virtualisation.
Roland’s presentation was followed by Colin Love from the University of London Computer Centre (ULCC). ULCC is a major co-lo site for HE and being able to charge their customers in relation to the power that they are using is of obvious interest to them.
Chris Cartledge, late of the University of Sheffield spoke next. He described the state of play with the university’s data centres. The picture is one that is probably familiar to a lot of people in the sector. This means plant and facilities that were built before people got really interested in data centre energy use, with poor air circulation and handling with mixing of the air flows.
Through all this discussion we keep coming back to a number of what are becoming familiar themes. One is the variable nature of metering – what do we meter and how? Do we meter at the power distribution board or at the rack or the individual box? What about metering embedded in the box itself?
Sheffield (with the help of funding from Salix) are working to make things better. A hot aisle is being introduced in one of the data centres, with a cold aisle in the other there there is no ceiling void. Racks will be sealed to prevent air mixing.
Sheffield is also working with the EU Code of Conduct for Data Centres. The University now includes the cost of energy in some VFM calculations.
“By metering you are not changing anything!” The invetsment in metering only makes sense when you do something with the information.
After coffee and biscuits Michael Rudgyard from Concurrent Thinking spoke. Michael’s background was in HPC, and he has come to his current approach from a belief that a data centre needs to be holistically managed and monitored. He said that PUE is a good metric for planning a data centre, but is not much use for monitoring a DC. What is more important is how much of data centre energy is actually being used to do useful computing work? Should we be looking to develop a metric like “ITUE”? This could be on the basis of page impressions/watt, database transactions/watt, widgets sold/watt.
He identified a number of steps that can be taken. One is to identify and to eliminate servers that are doing nothing useful and making sure that equipment is up to date working properly. Another is the dynamic orchestration of virtual machines based on environmental, power and IT usage constraints. It’s also worth looking at active power management of servers during low utilisation periods.
Turning to what he termed “The Problem of Scale” Michael highlighted the drive for consolidation in organisations and across the sector with Cloud computing. Big cloud providers are driving efficiency by building their own servers and removing redundant equipment and writing their own software to be more efficient.
He went on to talk about some of the features of the two products that Concurrent Thinking are bringing to market. This includes Concurrent Command and Concurrent Control – more details here: