Previous | Next --- Slide 20 of 40
Back to Lecture Thumbnails
Xelblade

The capacity should be just large enough to fully encompass some integer number of working sets since you get the greatest decrease in misses per unit capacity immediately preceding those points. (A capacity level right after the sections of steep dropoff on the graph.) Basically, fitting an entire "section" of data in the cache is important so you don't have to remove any data that you will reuse immediately and often.

Xiao

Elaborating on @Xelblade's explanation, from a cache designer's perspective it is important to maximize perf/cost efficiency (transistors do not grow on trees). Therefore when choosing adequate cache sizes, the designer will more often than not choose the points right after the drop-off happens, since these points produce the most noticeable change in data traffic.