- GreenSKU redefines server design for sustainable cloud computing efficiency
- Energy-efficient GreenSKUs address the rising carbon costs of cloud computing
- Microsoft leads the way in reusing server parts to reduce emissions
Servers consume large amounts of energy while operating and generate significant carbon emissions during their manufacturing.
To address this challenge, researchers at Microsoft, Carnegie Mellon University, and the University of Washington have developed “GreenSKU,” a sustainable approach to server design.
Unlike standard practices where components are discarded after 3 to 5 years of use, the GreenSKUs framework focuses on reusing parts such as random access memory (RAM) modules and solid state drives (SSD) from servers out of service.
Backwards Compatibility Unlocks Reuse
The foundation of GreenSKU lies in taking advantage of backward compatibility. Advances in technology, such as Compute Express Link (CXL) controllers, enable compatibility between different generations of memory modules, such as DDR4 and DDR5, making it possible to reuse older components without a significant loss in performance.
To ensure reliability, the researchers created a framework that evaluates components for reuse, identifying parts that will not cause unacceptable drops in performance or excessive power consumption.
The team’s efforts extend beyond hardware, introducing a layer of software to further refine performance that determines which computing tasks are best suited for GreenSKUs compared to standard Azure servers.
GreenSKUs feature low-power AMD Bergamo processors along with repurposed RAM and SSDs from obsolete servers, helping to reduce emissions linked to hardware manufacturing.
According to the Association for Computing Machinery, the carbon footprint of cloud computing is set to grow significantly, potentially contributing 20% of global emissions by 2030. Currently, cloud operations consume about 3% of the global energy annually.
The GreenSKUs project aligns with these efforts and researchers estimate an 8% reduction in embodied and operational carbon emissions for servers using this approach. At scale, the technique could reduce global carbon emissions by 0.1% to 0.2%, a reduction comparable to emissions from total smartphone use in the United States.
However, reusing older components is not without its challenges. The reused RAM modules introduced latency and reduced memory bandwidth, which the researchers overcame with memory pooling techniques. Similarly, slower read/write speeds on SSDs were addressed using RAID striping.
However, not all limitations had solutions. For example, AMD Bergamo processors, while power efficient, had reduced cache performance. The researchers built these trade-offs into their broader framework to ensure that each server configuration could still meet Azure requirements.
Through IEEE Spectrum
