Intelligent solutions to the challenges of virtualisation in the server room

Organisations generate masses of data on a daily basis, which needs to be stored somewhere accessible so that users can access it to do their jobs.

June 7, 2012

Robert Brand, UPS and Infrastructure Product Specialist at Drive Control Corporation

Organisations generate masses of data on a daily basis, which needs to be stored somewhere accessible so that users can access it to do their jobs. Storage requires servers, but given the sheer volumes of data in today’s world, the number of servers needed and the physical space they occupy has become a problem.

This has led to the evolution of the blade server, a technology that allows one physical machine to house multiple servers. This virtualised server room enables organisations to house more data in less physical space. However, by solving the single problem of space, the virtualised server room creates five more challenges that must be addressed to keep high density server deployments up and running optimally.

The first problem, which is the heart of the matter and leads to all of the other challenges, is the fact that each rack in a high density virtualised server room can use up to ten times as much electricity as a standard server rack. This is not because blade servers are less energy efficient than standard servers, in fact the opposite is true. However in such a virtualised environment, one physical server carries the workload of multiple virtual servers, which means that each machine has to work far harder to deliver on performance. This means that power consumption per rack is a lot higher, leading to increased utility costs and an increased carbon footprint.

Increased power consumption also means that the peak load in the server room is higher than previously, which necessitates larger capacity UPS solutions to enable the servers to perform a stately shut down or to continue operating in the event of a power failure.

Added to this, since each blade server is consuming more power to run at performance capacity, it correspondingly generates a lot more heat than a standard server. This effect is multiplied by the number of blades in a rack, so increased power consumption along with higher density also means that each server rack generates a large amount of heat. The increased capacity UPS solutions also generate more heat than previously, putting added pressure on cooling systems.

Sensitive computer equipment does not function well at high temperature, and must be cooled at risk of overheating and failing. When it comes to blade servers however, cooling the ambient temperature of the room is no longer enough. Blade servers require directed cooling, with fans that blow directly onto the face of each blade. Directed cooling using traditional methods, with the air conditioner mounted on the wall or ceiling is not the most efficient method of cooling, since there is inevitably a certain distance that the air must travel from the cooler before it reaches the server, so the air will warm up during this journey, decreasing cooling efficiency.

Blade servers are highly sensitive, requiring a constant, clean supply of power and efficient directed cooling. These challenges are compounded by the fact that blade servers are also very expensive pieces of equipment, so if power supply and cooling is not addressed and a server fails, organisations will be out of pocket for large sums of money along with the inconvenience of a non-functional server.

The challenges presented by the high density virtualised server rooms require intelligent solutions that address problems while optimising energy consumption to minimise carbon footprint and costs associated with electricity usage.

A modular, in-rack UPS solution solves certain of these problems. Using a modular UPS, the solution can to be right-sized for current needs, thus minimising wasted energy consumption resulting from a UPS that has been sized to accommodate future growth. As and when growth occurs, new UPS modules can easily be added to cater for this. These types of solutions also sit in the data rack, minimising their physical footprint and enabling them to benefit from the same cooling solution as the servers.

When it comes to cooling, in-row coolers are the intelligent solution to maintaining optimal server temperature while optimising energy consumption. These cooling systems sit in between the data racks themselves, delivering cold air directly onto the servers and drawing the hot air off them. This hot air is then drawn back into the cooling system and recycled. This ensures that the hot air does not cycle around, raising the air temperature and decreasing the effectiveness of the cooling.

However, blade servers do not have a constant power requirement – booting the servers up, idling and usage all create different power loads and different heat levels and cooling requirements as a result. Having a cooling system that constantly cools as if the maximum power load were being used is highly inefficient and unnecessarily wasteful. Using intelligent in-row cooling with built-in sensors will ensure that cooling is directed as needed to each individual server to maintain optimal operating temperature, accelerating and decelerating depending on how hard the server is working. In-row cooling solutions are also modular, so they can be added on when necessary as the capacity of the server room grows.

The benefits of intelligent in-row cooling are manifold. Firstly, temperatures can be kept at the ideal level for each and every server, with cooling directed at the face of each blade. This ensures that servers will operate effectively and extends the lifespan of the servers themselves, saving organisations money. Cooling systems also do not have to work as hard to cool servers, reducing wasted power and lowering energy consumption associated with cooling.

Addressing the challenges presented by the virtualised server room requires solutions with the intelligence to reduce wasted energy, lower total energy consumption ensure that servers perform to their peak capacity.