Opinion - Grids can be green
David Wallom is the technical manager of the Oxford e-Research Centre at the University of Oxford and chair of the UK e-Science Engineering Task Force. He is currently working on a JISC-funded Low Carbon Information and Communications Technology project.
Can grids really be green?
Grids have been designed to provide collaborating researchers with resources from many different physically disparate organisations. This has the advantage of ensuring that high-cost resources-in terms of initial purchase price and running costs-are operated as near to maximum capacity as possible.
This means that, through correct management of related resources, grids have the potential to effectively lower an organisation's overall carbon footprint.
Disposing with pretence
However, we should also recognize the natural resources consumed to build these computing systems in the first place, many of which are toxic.
After what is a very short lifetime, these systems are dismantled and "recycled" for their parts, often in developing countries, causing local health and environmental problems.
As part of the global push for a more ecologically sound footing for society, we must start to consider the longer-term usage of resources. We must calculate the total cost of ownership over the lifetime of a system in a much broader way, including the cost of natural resources, from construction through to dismantling after completion of its natural life.
This new method of considering "total cost of ownership" is especially important when comparing a desktop computer to a dedicated server system, located within a fully serviced machine room.
In this latter scenario the computer room uses large amounts of energy to keep the collection of systems cool. Due to the density of computers, the level of cooling required can be such that an equal amount of electricity is consumed by the cooling system as by the computers themselves.
To this end, cycle stealing from desktop systems must be used where possible. Cycle stealing involves the use of machines that are normally used for other purposes, such as office desktops or student teaching labs. It has been shown to give benefit in two distinct ways.
Firstly, cycle stealing enables higher throughput within large cluster resources since individual single-CPU tasks are removed and the systems are turned over to the specific types of computational task for which they were designed. Secondly, cycle sharing ensures that machines that might otherwise have been left turned on overnight and generally been a drain on institutional resources are able to make considerable contributions to research.
In the longer term this will hopefully mean that fewer systems will be built in the first place, which will ensure that the valuable and rare materials that are needed to build PCs will be used at a slower rate.
Further new advances within the machines themselves have enabled remote control of grid systems for power up and down. This means that systems will only consume resources during those periods when they are actually doing useful work. These types of systems will need to be intelligent enough to minimise switched-on time while ensuring that the physical lifetime of the hardware isn't compromised.
This type of approach can hopefully minimise the total number of new computers being purchased within a particular organisation, and hence reduce the production of e-waste, which will have follow-on benefits not only for the developing countries that process much of this waste, but also for our global environment.
- David Wallom, Oxford e-Research Centre, University of Oxford.