Notes on the paper Khan et al (2018) RAPL in Action: Experiences in Using RAPL for Power Measurements. Advantages and limitations of the Running Average Power Limit (RAPL) interface.
Removing the aggregation spine resulted in 5x higher speed and capacity, and 41% reduction in power.
Developers need metrics that are consistent and easy to optimize, but carbon intensity varies in time and space. So when is it a useful metric for developers to consider?
Firefox 104 introduces power consumption measurement in the profiler. What does this mean for analyzing website energy consumption?
Paper notes – Mitigating Curtailment and Carbon Emissions through Load Migration between Data Centers
This is a good paper that makes valid points about the possibilities of migrating flexible IT workloads, however it makes classic assumptions I see in most papers that discuss this topic.
Which languages are the fastest and most energy efficient? The simple answer is: C, C++, Rust. The accurate answer is: it depends.
Website carbon calculators are not very accurate, especially if they only use data transfer as the metric.
Is it a good idea to make predictions about future energy consumption? Even mature public businesses typically only issue earnings guidance for the next quarter. Maybe it’s because predicting the future is hard?
Is cryptocurrency / Bitcoin bad for the environment? In 2019 crypto consumed 70-90 TWh of electricity globally, with 60-70 TWh of that from Bitcoin mining. What that means for carbon emissions depends on where the mining happens.
Why isn’t carbon aware workload scheduling more common? Data center level scheduling is infeasible, so what are the opportunities for developers to implement more granular functionality?