Tweet
I notice the linux_perf module says this:
// For now the current implementation is just copying the perf output for the last test run, but rather easily could be adapted to take average of all test runs, etc
Supposing I want to maintain statistics for every run, how would I do that? Not just an average, but the full dataset of the value recorded for each run. For example, suppose I want to see the L1 miss rate: I need to record the number of L1 hits and L1 misses for each run, which makes two datasets, and then at the end compute the miss rate for that run, making a third dataset. I don't want to just record the average in the results, because maybe there's an outlier that's distorting the results. With the full numbers I can then export them from the results file to JSON.
I suppose the wider question is how do test results objects work? If I understood the API maybe I could record the results with it successfully, but I can't really follow pts_test_result*.php without knowing where to start.
// For now the current implementation is just copying the perf output for the last test run, but rather easily could be adapted to take average of all test runs, etc
Supposing I want to maintain statistics for every run, how would I do that? Not just an average, but the full dataset of the value recorded for each run. For example, suppose I want to see the L1 miss rate: I need to record the number of L1 hits and L1 misses for each run, which makes two datasets, and then at the end compute the miss rate for that run, making a third dataset. I don't want to just record the average in the results, because maybe there's an outlier that's distorting the results. With the full numbers I can then export them from the results file to JSON.
I suppose the wider question is how do test results objects work? If I understood the API maybe I could record the results with it successfully, but I can't really follow pts_test_result*.php without knowing where to start.