Assessing the balance of power for data centre operators
Achieving the lowest possible power usage effectiveness (PUE) rating should not be the only objective of a data centre operator. After all, the most energy efficient data centre is one which has lost all power! Furthermore there is a considerable grey area around the method of PUE calculation.
Effectively consolidating space and power constrained legacy data centres into more energy efficient ‘PUE-friendly’ environments ultimately requires their migration into modern facilities. These can offer the space, power and infrastructure necessary for supporting future compute requirements over the medium to longer term; five, ten, even fifteen years.
But for a realistic data centre PUE to be calculated it should include the power consumed by offices, general lighting, security systems and so on. As a minimum it should include transformer and UPS losses. Ideally it should be measured over a 12 month period or if calculated then it should be based on the worst case conditions (i.e the hottest day of the year) – even so many companies will look at the energy consumption of the air conditioning units under ideal conditions and only include this in their calculations.
Without doubt the PUE can be improved by stripping out levels of resiliency; N+1 is more efficient than N+N for example, but some customers will demand the reliability of N+N regardless of PUE. Reliability is still king and this necessitates resilient energy supply which must not be unduly compromised in pursuit of the ultimate in PUE ratings.
While a PUE of 1 is not possible as this would mean that no power is consumed other than that used by the IT Load, it is feasible in a modern facility to get close. The processes and procedures to lower the PUE should focus particularly on reducing power for cooling. This will require a combination of close-coupled CRACs, hot and cold aisle containment, higher data hall temperatures and, assuming sufficiently low ambient temperatures are available, fresh air cooling.
Although the use of fresh air cooling drastically reduces the energy consumption there are some losses which cannot be avoided. For example, getting electricity from the incoming feed to the rack wastes power due to transformer losses, cable resistances and UPS inefficiencies. Even the fresh air cooling will consume some energy due to the fans needed to circulate the air. Furthermore, regardless of cooling requirements, best practice dictates that the data centre fresh air is changed regularly.
When taking steps to reduce power consumption, server virtualisation is essential. An average server used to be run at 10% to 15% of capacity but it is now possible to virtualise new servers to run 20 to 40 virtual machines.
But organisations must consider the power implications for the type of rack hardware used for running the required applications or they may be forced into using more racks and more power than actually necessary. There is a common misconception that running low density racks instead of higher density ones will be less costly when it comes to power but the reverse is actually the case.
Running fewer high density racks than lower density ones will yield a lower total cost of ownership because they have far superior compute capabilities while using significantly less data centre resource; switchgear, UPS, power, cooling towers and pumps, chillers, lighting and so on.
The problem is that the latest more efficient higher density racks consume over 5kW and a growing number more than 10kW: few data centres can actually supply this level of power per rack today and this problem is only going to get worse.
Integrated energy management
Central to optimising overall data centre energy efficiencies must be an advanced energy monitoring and management platform, capable of integrating the building management system, the electronic management system, PDUs and SCADA; data centres have historically used disparate systems which is considerably less efficient.
With an integrated energy management system there is the means to closely monitor energy usage, highlighting any areas of concern where consumption is running at unexpected levels. This can then be addressed quickly and efficiently, ensuring a better service for customers of the data centre and potentially saving the operator and its customers thousands of pounds through reduced electricity costs and of course minimising the environmental impact of their operations.
- » How to improve supply chains with machine learning: 10 proven ways
- » NASCAR moves onto AWS to uncover and analyse its racing archive
- » Why the real multi-cloud motivator is choice - rather than lock-in
- » Google Cloud looks to Looker for greater data analytics – but with the multi-cloud focus
- » How to prevent AIOps from becoming just another cog in the machine