Virtualisation and the data centre landscape

by

21 May 2015

Virtualisation is continuing to shape the data centre landscape, more so in Singapore than in many other parts of the world due to its position as the data centre hub of the region.

During a curtain raiser for Singapore Datacentre Week 2015, Franck Petit, Southeast Asia regional manager, DCD Group, presented some global findings from DCD Intelligence which indicated that the percentage of virtualised systems or cloud located on-premise grew slightly from 15 per cent in 2014 to 17 per cent, while the percentage of virtualised systems of cloud located externally grew from 10-12 per cent.

At the same time, physical racks located in-house slipped from 54 per cent in 2014 to 51 per cent in 2015, while the percentage of physical racks located with third-party facilities edged slightly down from 21 per cent to 20 per cent.

While these percentages are still relatively small, the shift looks set to accelerate in coming years, with virtualisation and cloud emerging as the second top driver for data centre investments in 2014-2015 (49 per cent), slightly behind the need to reduce capital operational expenditure (51 per cent). More significantly, the figure for virtualisation and cloud represented a 19 percentage point increase from the 30 per cent reported in 2013-2014. The only other investment driver that saw such a huge increase was the need to improve sustainability, which went from 18 per cent to 37 per cent.

The trends in Singapore are similar, but they are happening much faster and on a larger scale due to the country’s position as a regional data centre hub, said Petit.

Virtualisation also figured prominently in a panel discussion on how enterprises were innovating in terms of their IT strategy. Manik Narayan Saha, chief information officer, SAP , said one of the big IT discussions in the past couple of years has been around how enterprises can maximise utilisation of capacity that they have in-house or are buying from a service provider. “Virtualisation has led to the fact that you can increase virtualisation in multiple dimensions, be it compute, storage or networks,” he pointed out. “Organisations are also able to offset utilisation in certain units with non- peaks and troughs in the workload.”

For SAP, about 80 per cent of its internal servers are virtualised, and the company has about 70,000 virtualised servers globally, said Manik.

Echoing the importance of virtualisation and cloud as part of the enterprise IT strategy, Abhishek Parolkar, Head of Technology, PropertyGuru, said, “From our infrastructure perspective, everything is cloud. Virtualisation is integral in everything that we do. We want to fully utilise the investments that we have made.”

The panel also discussed virtualisation in the context of energy efficiency and energy consumption – significant issues for Singapore given that data centres account for about 7 per cent of the country’s total energy consumption, compared with the average of 2 per cent in United States and Europe.

Ed Ansett, chairman and founder, i3 Solutions Group, highlighted the counterintuitive fact that virtualisation, despite leading to a consolidation of IT resources and generally lowering electricity bills, actually has the effect of worsening the power usage effectiveness or PUE, which is one of the common indicators of data centre energy efficiency. This is because if power and cooling systems are not adjusted accordingly, the proportion of utility power that these systems consume will constitute a higher percentage of the overall data centre energy use even as IT workloads fall.

Setting aside such “outdated” measures of energy efficiency, Ansett said the real opportunity for reducing data centre energy consumption lies in what can be done from a software point of view. “The way in which organisations code, they are generally not thinking about how they can make the code more efficient,” he said.

For example, one way to reduce the energy consumed by software is to invoke sleep states and make acceptable compromises around wake up time. Generally there are three types of applications, he said. The first of those are applications which run according to a predictable timetable, for example, weekly or monthly. These could be put to sleep and woken up when needed. Then there are the applications in the unpredictable category. Of these, there are some that are very time sensitive, for example, applications for high-frequency trading which cannot be put to sleep at all. However, for other applications which are not so time sensitive and where it does not rally make a difference whether it takes another millisecond to wake them up, there is a huge potential for invoking sleep states as a way of driving greater energy efficiency, said Ansett.

In fact, as Manik of SAP pointed out, virtualisation was one of the ways in which systems could be put to sleep.  It allows virtual machines to be archived to disk, so they do not consume any energy if they are not in use.

“It is a huge opportunity for us,” he said. But the challenge lies with the migration path, and the need for IT to engage with the business more in these conversations. “For example, if you have a business unit that runs a report once a month, and you ask to shut down the system if they are not using it, it takes a lot of convincing too make them understand that they don’t need it for the other 29 days of the month.”

Joining in the discussion, Wong Ka Vin, managing director, 1-Net, said the low hanging fruit when it comes to lowering energy consumption in the data centre is for enterprises to try to arrive at a more accurate gauge of their data centre requirements. “Today, 90 per cent of customers do not get it right,” he said.

Enterprises may come up with a figure, say, 10kW, by adding up the power requirements of individual components in an IT system, and multiplying this by the number of systems they need, while factoring in several buffers along the way. “By the time this reaches the data centre provider, we are provisioning a lot of copper which you are not using. If we can save 30 per cent of that copper, that’s green IT.”

According to Wong, the biggest challenge facing data centres is to map their resources to make sure that they are used optimally. “If I have the ability to understand the arrival rate, and the power consumption of the equipment that is going to arrive, then I can provision the equipment accordingly and can give an optimal price. If not, I have to hedge capex for unpredictable loads.”

“Over the next couple of years, the role of the centre will continue to be about providing power and cooling, but we have to be a lot smarter and a lot closer to our customers,” he said.