In my research group we're sometimes benchmarking parallel systems (databases with parallelizable queries). Whenever there's a student thesis, I force them to explain Amdahl's Law in their thesis, and contextualize their work regarding this Law.
From this most students correctly get the idea that they have to analyze the parts of the workload seperately, as their parallelization behaviour is not uniform.
For example, loading the data from a spinning disk - given one disk - behaves differently than, for example, the evaluation of a query once the data has been loaded.
From this most students correctly get the idea that they have to analyze the parts of the workload seperately, as their parallelization behaviour is not uniform.
For example, loading the data from a spinning disk - given one disk - behaves differently than, for example, the evaluation of a query once the data has been loaded.