I don't understand why the speedup here is described in terms of work per unit time, but for time constrained scaling, it is simply described as work in a fixed execution time. Can we think about these in the same way? Or are they different?
@ok Going by dimensional analysis, the two formulas would seem the same (just cancel out "per unit time")--but the phrases used in the equations actually have somewhat different interpretations for each scaling type, so you can't really understand the whole concept from just the formula. On this slide, "work per unit time on P processors" means "the most work that can be done per unit time on P processors without overflowing main memory," whereas in time-constrained scaling we could use disk memory all we want to increase our throughput as long as the accesses don't push us over our time limit.