Poor heat and cooling testing costs data centres millions

Published on 19th December 2014 by Dave Wolfenden

It doesn't matter whether it is a new data centre or a refurbishment if getting the cooling balance wrong means throwing money away. The problem for many data centre builders is that irrespective of new build or refurbishment, they often have no idea what the real heat load will be in the data centre. As a result they do limited testing which is often poorly designed and their clients eventually end up paying the price.

What can you do to make heat load testing more effective and reduce the cost of cooling? Dave Wolfenden, director, Mafi Mushkila gives us some advice.

Tests must have some basis in reality

One of the biggest problems with testing is making it representative of the end use of the product. It doesn't matter if it is a car, a washing machine, a laptop computer or a data centre. If the tests have no basis in reality, then not only is the time and money spent on them a waste, but the false impression that they give of efficiency means that future waste is completely undetected.

In the data centre, one of the most common ways of testing the heat load is to introduce a heat source. There is nothing inherently wrong with this provided that it is done in a real-world way. For example, introducing a heat source of 20kW, 40kW, 60kW or higher might seem like a good way to discover what the cooling system can handle. However, if there is no equipment other than the heat source in the room, all that is being tested is the ability of the cooling systems to deal with hot spots rather than normal computer load.

One of the hardest things to get from the end-user client is expected-loading inside the data centre. While customers put down numbers in their specification, they are often guestimates rather than realistic numbers. With long lead times for data centre construction at present, it is also possible that the hardware originally destined for the data centre may have changed.

The solution is to get a customer to provide a range of value for each data centre hall giving the expected lower end, the maximum expected load and, where possible, some idea of the type of systems to be installed. With the latter, it is then possible to place variable heat sources around the data centre in order to best match how heat will be generated when systems are running.

How to make testing more effective

There are a range of actions that can be taken to make testing more effective and it doesn't matter if this is a new build or a refurbishment. The key is to get infrastructure and load emulators into the hall and configure them to be as representative of reality as possible.

Ten things that can be done to make testing more effective include:
Racks and cable trays are the minimum type of equipment required. To make the cable trays more effective, tape off some of the ducts to represent different densities of cable load.

Make sure that the racks are all properly blanked off to prevent air mixing and arrange them in a similar configuration to the end user requirement.

Not all infrastructure will be rack mounted so add in additional components to represent the type of equipment often found in the data centre.

Make sure that the expected means of input and exhaust air are accounted for along with the common types of air flow interference and arrange the room accordingly.

Place multiple sensors in each rack and row to get a granular view of air from the ground to the ceiling.

Use multiple heat emulators per rack and place them where the load will occur. For example, if the rack will have multiple switches mounted at the top of the rack, place a heat emulator there. Similarly, if there are going to be blade servers which generate large amounts of heat at the bottom, place a larger heat emulator at the bottom.

Don't test with a single heat load. Vary the loads from the emulators across the racks and throughout the day to make this representative of normal IT workloads. Focus on the edge cases such as peak logon and backup times where systems will be under high stress loads.

If designing or refurbishing multiple data halls, create or buy movable racks that will hold the heat emulators. With hardware refreshes taking place every 3-5 years and data centres having a life of up to 25 years, it makes sense to invest in equipment that will support a rolling program of refurbishment.
 

Computational fluid dynamics
Invest in or hire specialists in computational fluid dynamics (CFD) who will be able to see how airflow moves as you alter the heat load. This will quickly identify where there is a risk of hot spots that are not easily cooled and provide information as to where certain types of equipment with high heat load can/cannot be located.

Ensure that you create a set of baselines for the different types of test load. This can then be used to compare against ongoing readings from the sensors once the data centre has been commissioned. Evaluating actual vs projected heat is a good indicator of future problems and potential energy waste.
 

Using third party testing

These ten steps are not an exhaustive list and the choice of whether to use some or all will depend on budget and availability of equipment. In most cases, it is possible to bring in a third party to do the testing and they will supply more of the equipment required to make the testing realistic.

Experienced testing vendors will be able to advise on how best to create realistic baselines for the types of workload to be run. They will also help create the processes that then compare tests, projections and actual heat and cooling figures gathered by sensors. While these figures are significant in maintaining future costs and efficiency they are also indicators of non IT related problems.

While the focus on heat and cooling is often around the hardware, poor housekeeping practices such as a build up of old cabling or a failure to maintain cable standards can have a disproportionate impact of the effectiveness of cooling. While some of these issues are easily spotted by walking around a data centre, the sensors showing an increase in heat and regular use of CFD to test airflow will also indicate hidden problems, especially where infrastructure is under the floor or above the ceiling.

Conclusion

It is not difficult to create a valid heating and cooling test environment but it does require planning. The biggest issue is often the communication between the end-user client, the contractor creating/building the data centre as well as the test team. In many cases, while there is a contract and a handover process, there is little real communication over future use.

One reason for this is that those commissioning the data centre are not part of the IT team. There is still a disconnect between facilities management and IT and this will always create opportunities for money to be wasted. Another reason is commercial sensitivity where corporate customers want to prevent competitors gaining an understanding of their future data centre requirements.

Irrespective of why customer and contractor do not talk to each other, both parties must take responsibility for historical poor practices in testing data centres. Solving the problem is not hard, will save money on energy costs and is an important contributor to any corporate environmental audit.

For more information visit www.heatload.co.uk

CONNECT WITH US

Sign in


Forgotten Password?

Create MyDCD account

Regions

region LATAM y España North America Europe Em Português Middle East Africa Asia Pacific

Blogger

Whitepapers View All