When a cache line is brought into the last-level cache, another line must be evicted to make room for it. When lines in active use are evicted, a performance problem may arise from continually rotating data back into the cache. This metric measures the percentage of all replacements due to each row. For example, if the grouping is set to 'Function', this metric shows the percentage of all replacements due to each function, summing up to 100%.
This row is responsible for a majority of all last-level cache replacements. Some replacements are unavoidable, and a high level of replacements may not indicate a problem. Consider this metric only when looking for the source of a significant number of last-level cache misses for a particular grouping. If these replacements are marked as a problem, try rearranging data structures (for example, moving infrequently-used data away from more-frequently-used data so that unused data is not taking up cache space) or re-ordering operations (to get as much use as possible out of data before it is evicted).