based on implementation described in Ketan Shah 2010
We implemented (unoptimized) two version of LFU Replacements described in STEFAN PODLIPNIG 2003
"Perfect LFU counts all requests to an object i. Request counts persist across replacements. On the one hand, this assures that the request counts represent all requests from the past. On the other hand, these statistics have to be kept for all objects seen in the past (space overhead)."
"With in-cache LFU, the counts are defined for cache objects only. Although this does not represent all requests in the past, it assures a simpler management (less space overhead)."