To compare the two techniques, we apply both the DEMOTE and the PROMOTE techniques to two popular single-level caching policies. While LRU is the most fundamental caching technique (variants of which are widely used [26,13,11,31]), ARC is arguably the most powerful single-level caching algorithm [25]. A multi-level cache performing the same as a single-level ARC cache represents the most powerful application-independent, multi-level caching policy.
DEMOTE-ARC is implemented by maintaining the 
and 
lists
as global lists, divided among all caches in proportion to the size of the caches [12].
The aggregate hit ratio is precisely equal to that obtained by a single cache of the
aggregate size implementing ARC.
This is the strongest multi-level caching contender we can devise.
DEMOTE-LRU is implemented as suggested in earlier work [33], also depicted earlier in Figure 2. PROMOTE-LRU and PROMOTE-ARC are implemented as explained in Section III.
For completeness, we also compare the performance of LRU, which is defined as the simple Least Recently Used (LRU) policy implemented in each cache within the hierarchy. There are no inclusion or exclusion guarantees since each cache behaving as if it were the only cache.