The Energy Efficient Computing Research (EECR) programme at the Hartree Centre is investigating ways of improving the efficiency of its data rooms.
Our aim is twofold: To increase the efficiency of our data rooms – maximising our utilisation while minimising our energy consumption, and also to develop our data rooms into a model of evolving best practice.
This is an interesting and useful challenge as many of the optimisations employed in an HPC data centre are also applicable to commercial data centres with potential beneficial impact for the economy and the environment. It’s fair to say that the growth in number and scale of datacentres around the world and increasing energy sustainability concerns mean that optimisation at all levels is vitally important if we are to reduce the impact of the carbon foot print from this growing sector.
I’m sure many data centres, like the data rooms at the Hartree Centre have grown and evolved over a number of years. The hardware and supporting infrastructure provide varying levels of compute and efficiency all of which need to be managed and optimised.
“Understanding how each element of this infrastructure is performing and its wider impact of efficiency and the risks and costs associated with it is important.”
Understanding how each element of this infrastructure is performing and its wider impact of efficiency and the risks and costs associated with it is important. This includes:
- Power usage – of computing hardware and supporting infrastructure. The overall efficiency of a data centre is generally measured via its Power Usage Effectiveness (PUE), which is the total amount of power consumed in total divided by the power consumed by the compute and expressed as a ratio.
- Temperature – is also important, not to just understand if we are overcooling the rooms but whether there are hot or cold spots which could affect long term reliability and drive up costs.
- Humidity – avoidance and reduction of humidity are important considerations for data centre management.
- System utilisation – How much usage are we making of our computing resources compared with the total cost to run them? Do they offer good value for money?
When you have accurate data you can start to evaluate the data rooms, combining utilisation data along with power, temperature and humidity measurements provides useful information we can use to optimise operational conditions. It can also inform decisions around decommissioning and the deployment of new systems and potentially the design of a new and more efficient data centre in the future. Having this data available also allows more informed service planning, allowing a more robust comparison of on premise provision versus HPC cloud solutions.
In order to provide this level of monitoring and reporting the Hartree Centre has in place a research collaboration with Concertim Infrastructure Management to install a range of power, temperature and humidity monitoring equipment throughout our data rooms. This system has been designed to capture data from our existing systems as well as the new Power 8 machine and the planned Central Data Store. Over the coming months I’ll report back on progress and outline some of the planned changes we hope to implement and their impact.
You can find out more about our energy efficient computing programme in our animation.
Please let us know what you think in the comments below.