Under the Timing Issues header, the project description states:
"To get good timing results, you will need to repeat an operation (such as
the context switch) many times and divide the total time by the number of
repetitions."
I was asked about this and I wanted to post here so it could be discussed. I have not been dividing by the total number of times as the direction suggests. Reporting the minimum has been mentioned many times as a more appropriate choice for our results.
We reported the min,median,max and quartiles since the data was within some reasonable range of the median anyways. The min and max were outliers in most cases.
ReplyDeleteOn speaking with my reading group, Take context switching. I'm guessing the most accurate time for what it actually takes the OS to context switch is MIN because of the extra overhead our experiments introduced. For example, if waiting on a mutex was implemented as a busy wait, then you'd take the minimum time because it represented the run where the greatest fraction of time was spent context switching.
ReplyDelete