In our measurement docs we define a unit of work for carbon measurement as "X kernel events for Y minute(s)" which would be useful if we had an ideal "average events per minutes" that applied across workloads generally. A workload and a unit of time combine in a way that allows us to calculate both carbon emissions and embodied carbon.
The benchmarking plan for Falco is expected to be based on stress testing workloads. Stress testing will likely be based on a higher X per Y than an average workload. the output of such a benchmark would be useful, but more as a "energy per kernel event" calculation, lacking a time element. This can be translated into a timescale if we layer on an average events per minutes.
I'm wondering if there are already average events per minutes estimates floating out there? We could theoretically pick a number out of thin air, make it explicit, and suggest that future calculations be adjusted based on the reality of the project in question.
In our measurement docs we define a unit of work for carbon measurement as "X kernel events for Y minute(s)" which would be useful if we had an ideal "average events per minutes" that applied across workloads generally. A workload and a unit of time combine in a way that allows us to calculate both carbon emissions and embodied carbon.
The benchmarking plan for Falco is expected to be based on stress testing workloads. Stress testing will likely be based on a higher X per Y than an average workload. the output of such a benchmark would be useful, but more as a "energy per kernel event" calculation, lacking a time element. This can be translated into a timescale if we layer on an average events per minutes.
I'm wondering if there are already average events per minutes estimates floating out there? We could theoretically pick a number out of thin air, make it explicit, and suggest that future calculations be adjusted based on the reality of the project in question.