bizoCarbonReductionOpportunity2019#
- Biz19
Daniel Bizo. The carbon reduction opportunity of moving to Amazon Web Services. AWS, October, 2019. URL: https://d39w7f4ix9f5s9.cloudfront.net/e3/79/42bf75c94c279c67d777f002051f/carbon-reduction-opportunity-of-moving-to-aws.pdf.
Summary
The authors have done an extensive analysis of what a “typical” enterprise datacenter is, and how it is operated. They conclude that AWS is about 3.6 times more energu efficient, mostly becuase of more efficient equipment, and more efficient operations.
They have only analysed the use phase, not the manufacturing, distribution, or end-of-life phases.
Snippets#
Our results show that AWS’s infrastructure is 3.6 times more energy efficient than the median of the surveyed US enterprise data centers. More than two-thirds of this advantage is attributable to the combination of a more energy efficient server population and much higher server utilization. AWS data centers are also more energy efficient than enterprise sites due to comprehensive efficiency programs that touch every facet of the facility.
Note
Some reasons why AWS is more efficient:
Temperature control seems to be one source of difference. This paper claims that traditional datacenters are actively cooled with compressors, causing high energy consumption, whereas AWS uses evaporative cooling and servers that can operate on wide temperature bands.
Low electrical efficiency (below 80%) caused by out-dated equipment and inability to operate efficiently at low load.
The study shows that, on average, the target group of US enterprises tends to keep their servers for a little over four years before upgrading, although the average masks a wider spread: some keep a server for as long as seven or eight years while others say they refresh after less than three years as a target - a quite aggressive approach. This policy is not strongly correlated with company size or vertical.
About 93% of the enterprises in the sample reported that they use virtualization (the rest of the respondents did not know).
[…] very few companies consistently outperform their peers, indicating that even with initiatives to achieve best-in-class operational efficiency within an organization, they are not generally effective enough to raise all aspects of enterprise operations in line with best practices.
[…] the latest-generation Intel servers still show a factor of 2-3x difference between their peak efficiency point (roughly 70% load) and their light load (10-20%) range.
Data shows that virtualization and level of workload consolidation most strongly explain the difference in efficiency among enterprises because both markedly increase utilization levels. Perhaps more surprising is that the third-strongest factor that correlates with how well enterprises fare against their peers is speed of technology adoption, which is statistically slightly more influential than server lifecycle policy.
AWS also designs its own servers for maximum efficiency, while enterprises might give more consideration to other features such as hardware redundancy and expandability