- GreenSKUs redefine server design for sustainable cloud computing efficiency
- Energy-efficient GreenSKUs tackle the growing carbon costs of cloud computing
- Microsoft leads in reusing server parts to cut emissions
Servers consume vast amounts of energy while operational and generate substantial carbon emissions from manufacturing.
To address this challenge, researchers from Microsoft, Carnegie Mellon University, and the University of Washington have developed “GreenSKUs,” a sustainable approach to server design.
Unlike standard practices where components are discarded after 3–5 years of use, the GreenSKUs framework focuses on reusing parts like random-access memory (RAM) modules and solid-state drives (SSDs) from decommissioned servers.
Backward compatibility unlocks reuse
The foundation of GreenSKUs lies in leveraging backward compatibility. Advances in technology, such as Compute Express Link (CXL) controllers, allow compatibility between different generations of memory modules, like DDR4 and DDR5, making it feasible to reuse older components without significant performance loss.
To ensure reliability, the researchers created a framework that evaluates components for reuse, which identifies parts that won’t cause unacceptable performance declines or excessive energy consumption.
The team’s efforts extend beyond hardware, introducing a software layer to further refine performance that determines which compute tasks are best suited for GreenSKUs compared to standard Azure servers.
The GreenSKUs feature energy-efficient AMD Bergamo processors alongside reused RAM and SSDs sourced from defunct servers, helping to cut emissions tied to hardware manufacturing.
According to the Association for Computing Machinery, cloud computing’s carbon footprint is poised to grow significantly, potentially contributing 20% of global emissions by 2030. Currently, cloud operations consume around 3% of global energy annually.
The GreenSKUs project aligns with these efforts and the researchers estimate an 8% reduction in embodied and operational carbon emissions for servers using this approach. At scale, the technique could lower global carbon emissions by 0.1–0.2%, a reduction comparable to emissions from all smartphone usage in the United States.
However, repurposing older components isn’t without challenges. The reused RAM modules introduced latency and reduced memory bandwidth, which researchers overcame with memory pooling techniques. Similarly, lower read/write speeds in SSDs were addressed using RAID striping.
However, not all limitations had workarounds. For example, AMD Bergamo processors, while energy-efficient, had reduced cache performance. Researchers incorporated such tradeoffs into their broader framework to ensure each server configuration could still meet Azure requirements.
Via IEEE Spectrum