Geography helps HPC go green
In the final article in this series, Tom Wilkie and Robert Roe look how geography can make computing more environmental – and encourage smaller companies to benefit from HPC as well
The BMWi series cars are highly energy efficient – the i8 is a sports car capable of going from 0–100km/h (0 to 60 mph) in 4.4 seconds and with a top speed of 256 km/h (159mph). Yet because it is a plug-in hybrid, it consumes only 2.1 L/100km or 134.5 miles per gallon of fuel.
But the i8 has a further claim to environmental fame. BMW ran the computer simulations and designed the car using computers based in Iceland, thousands of miles away from its Munich base. Speaking at the High-Performance Computing and Big Data Congress, held in London on 3 February, Jorge Balcells, director of technical services for Verne Global, pointed out that the move has saved 3,500 tonnes of carbon and reduced BMW’s energy costs by 82 per cent.
As the earlier articles in this series have shown, strategies to improve energy efficiency include better software, newer processors, and better cooling. But geography is also being employed to reduce energy costs and improve the environmental impact of both HPC and commercial datacentres.
Two companies were offering cloud computing for HPC at the London meeting in February: Verne Global, at whose data centre in Iceland BMW is now carrying out its computing; and Hydro66 in northern Sweden. Their data centres are powered by renewable energy and a further attraction of locating in the far north is the low ambient temperatures that will lower the cost of cooling.
In enterprise computing too, the focus has turned to Europe. Last month, Apple announced that it would build two data centres in Europe, costing a total of €1.7 billion and each powered entirely by renewable energy. Lisa Jackson, Apple’s vice president of environmental initiatives said: ‘We believe that innovation is about leaving the world better than we found it, and that the time for tackling climate change is now. We’re excited to spur green industry growth in Ireland and Denmark and develop energy systems that take advantage of their strong wind resources.’
Although the announcement was thin on details, the two Apple data centres, each around 166,000 square metres are designed to capture excess heat from equipment inside the facility and conduct it into a district heating system to help warm homes in the neighbouring community. Although this is not a new idea to the HPC industry it requires investment and forward planning.
By locating the two data centres in Denmark and Ireland it will be possible to use evaporative cooling techniques thanks to the cool climate of the chosen locations. This is something that has been done previously in HPC facilities, in a report from last year Robert Roe discussed how the German SuperMUC system, housed in the Leibniz-Rechenzentrum (LRZ) uses simple ‘free-cooling’ equipment because, the air temperature rarely exceeds 35 degrees Celsius, even though Munich, with a continental climate, has hot summer weather.
But going further north improves the benefit from local weather. Verne Global bases its data centre in Keflavik, Iceland, providing access to renewable energy and low temperatures. Similarly Hydro66 has based its datacentre in Boden, Sweden, where the temperature can get to -25C in winter.
In 2013, Verne Global partnered with Datapipe, a managed services and infrastructure solutions provider, to offer customers its Stratosphere high performance computing (HPC) cloud platform delivered from Verne Global’s data centre campus in Iceland. The company's Stratosphere HPC cloud platform aims to provide on-demand scalability with performance and security benefits, as well as long-term, predictable power costs.
According to Verne Global's CEO Jeff Monroe, there are two leading constraints for HPC clouds and clusters: power availability and costs. Monroe said: ‘Together, Verne Global and Datapipe are meeting these challenges with the first truly green HPC Cloud for the European and North American markets.’
Hydro66 sources all its electricity from a 70MW hydro-electric power station just 500 meters away from its facility on the local Luleälven river. The result is zero CO2 emissions and, the company claims, the cheapest electricity in Europe.
Verne Global too draws on completely renewable, zero emission, geothermal and hydroelectric energy. However, speaking in London in February, Verne Global’s Jorge Balcells stressed that it had negotiated a long-term energy supply agreement. The company thus guarantees the provision of electricity at a low, inflation-protected rate for up to 20 years. As much of Europe faces rising energy costs, he sees this as a particular advantage of the facility’s location in Iceland.
Cloud-based HPC provides an interesting new aspect to data-centre design that was once only available to global companies like Apple. A high-speed fibre network can deliver the bandwidth required to move the large amounts of data required for HPC and thus enable organisations and companies to locate their data centres anywhere in the world that is connected up to such a network – as both Iceland and Sweden are.
Hydro66 has built in redundancy with four fibre pathways offering bandwidths from 1Gbps to 1Tbps. Speeds greater than 10Tbps are also available but require advance notice. Verne Global too has redundant, high-capacity, multi-terabit-per-second connections, including Farice-1, Danice and Greenland Connect. It claims a latency of only 20 milliseconds from both London and Paris and 41 milliseconds from New York. Danice links to Denmark and the Netherlands with a capacity of 5.1 terabits per second. It was chosen to reduce latency to Central and Eastern Europe, and also to provide proximity to connection points in Stockholm and Amsterdam.
Discussions at the Machine Evaluation Workshop, held in Coventry, UK, in December last year highlighted how conventional HPC and the cloud were converging, offering new solutions that can be scaled and tailored to the size of an individual workload. Cloud-based HPC suits those who either want to increase peak performance temporarily for a specific job, or a small company that needs to expand without buying the associated HPC infrastructure up front.
For smaller companies, the only viable way of accessing HPC is either to rent hours on an existing machine, co-locate with one or more businesses using a shared infrastructure, or make use of virtualised HPC. So delivery of HPC services via the cloud can be advantageous to smaller companies – and the energy efficiency is an added bonus. With an increased uptake of HPC from small and medium sized businesses, see a previous article on the SHAPE programme, there will be an increased need for new datacentres that can provide the required compute that smaller companies need.