I spend quite some time to understand this formula.
The way to look at this formula is (increase in work)/(increase in time). The underline assumption is that memory on each core remains the same.
This formula is a fair way to measure speedup. A naive way to measure it is simply to use (increase in work) to measure the performance by adding more processors. But this formula is not reasonable if we ignore the time. For example, if we get p times work done but the synchronization overhead leads to p times running time. We cannot say the speedup is p. It is actually p/p = 1. That's reason we use (increase in work)/(increase in time) to measure speedup.
I spend quite some time to understand this formula.
The way to look at this formula is (increase in work)/(increase in time). The underline assumption is that memory on each core remains the same.
This formula is a fair way to measure speedup. A naive way to measure it is simply to use (increase in work) to measure the performance by adding more processors. But this formula is not reasonable if we ignore the time. For example, if we get p times work done but the synchronization overhead leads to p times running time. We cannot say the speedup is p. It is actually p/p = 1. That's reason we use (increase in work)/(increase in time) to measure speedup.