[ https://issues.apache.org/jira/browse/CASSANDRA-11452?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15241025#comment-15241025 ]
Benedict commented on CASSANDRA-11452: -------------------------------------- Also, just to clarify, I'm not proposing a random _eviction_, just a random selection of who to compare against for _admission_ - the eviction candidate would still be the LRU. Thus the collision would always be removed within a short number of steps after reaching the LRU spot, and ordinarily rapidly after. It's also worth noting that a RNF whose average walk distance was only a little larger than 1 (so that it usually compared against the eviction candidate) would more than suffice - if the chance of each distance was 1/4 of the prior distance, the average walk length would only be 1.33, but it would still take only a few comparisons for the eviction to unblock, and a few more for multiple such collisions to be resolved. > Cache implementation using LIRS eviction for in-process page cache > ------------------------------------------------------------------ > > Key: CASSANDRA-11452 > URL: https://issues.apache.org/jira/browse/CASSANDRA-11452 > Project: Cassandra > Issue Type: Improvement > Components: Local Write-Read Paths > Reporter: Branimir Lambov > Assignee: Branimir Lambov > > Following up from CASSANDRA-5863, to make best use of caching and to avoid > having to explicitly marking compaction accesses as non-cacheable, we need a > cache implementation that uses an eviction algorithm that can better handle > non-recurring accesses. -- This message was sent by Atlassian JIRA (v6.3.4#6332)