Data centres consume a lot of power and generate a lot of operational data. A canny engineer at Google found a way to use the data generated during data centre operations to get more bang for buck for Google.
Google engineer Jim Gao decided to use his 20% time – the day per week many Google staff get to work on pet projects – to find a way to reduce the PUE (Power Usage Effectiveness).
Gao hit the books and started looking into machine learning. He used what he discovered to create models that predicted, to a high degree of accuracy, how lots of different factors interacted to alter data centre efficiency. His models are 99.6% accurate in predicting PUE.
By getting a better understanding of what is influencing PUE at any moment in time – Google monitors PUE every 30 seconds – the company can now able to allocate resources and move loads around so that they can reduce power use. Hit the blog post for a detailed explanation.
Better data centers through machine learning [Official Google Blog]
Comments