An interesting finding from KPMG reveals that about USD 12 trillion will be spent on energy in the next 20  ears, equaling the spending of the past 100 years. Since server power efficiencies are not keeping pace with performance efficiencies, this escalates energy costs.

Having studied such astonishing facts, Dr Preeti Ranjan Panda, Associate Professor, Department of Computer Science & Engineering, IIT Delhi, decided to start a research on the subject. Dr Panda is working on a project to reduce power consumption in the data center by 50 percent. Dr Panda applied for and won a research grant from IBM through IBM’s Shared University Research.

Power Saving
There are various energy inefficiencies in the servers at the processor and memory levels. These inefficiencies can be eliminated at the architecture and OS-levels to bring down energy consumption in computing environments.

Says Dr Panda, “As processors get smaller, heat dissipation increasingly becomes a problem. Moreover, the increasing amount of main memory puts further pressure to the thermal envelope of a system. These power guzzling servers also require a similar amount of power to cool them. While there are various power management tools built into the OS to put the disks in standby mode etc to reduce power, a lot of energy still gets wasted. An idle system consumes about 50 percent of its peak power on an average. If we have better process scheduling techniques, 50 percent energy can theoretically be saved.”

There are however certain challenges involved. “In order to achieve this kind of power saving, you should be able to predict the workload and shut down systems that are not in use. If all idle systems are shut, any increase in demand would lead to slower performance since even automatically switching on a system would take some time,” he adds. Thus, there is a trade off between power and performance which this research is trying to minimize.

Dr Panda along with some of his research students are trying to see through software simulations to find out how accurately we can predict the workload in order to ensure both good performance and better power efficiency. “The system should be automatically able to predict which systems can be shut down and which  systems need to be fired up,” explains Dr Panda.

While initially IIT Delhi is trying to see how it can best utilize the power consumption in the IBM cell processors, the research would be applicable to all server processors including Intel’s x86 processors. “Cell processors are powerful processors. We feel they can be much better utilized with some software tweaking,” says Dr Panda.

An estimated Rs 10 to 12 lakh per year for a span of at least three years is expected to be spent on the research and funded by IBM.

With such research taking place in India, the country could become a leader in energy-efficient technologies. It may not be too long before other countries begin adopting our technologies rather than vice-versa.