[ https://issues.jboss.org/browse/ISPN-8397?page=com.atlassian.jira.plugin.... ] Sebastian Łaskawiec commented on ISPN-8397: ------------------------------------------- Current setup: * Container Size - Xmx (100 MB) * JVM overhead (200 MB) * 30% of Container Capacity (measured empirically, lower value caused the container to crash) Test details: * Test was performed with 1K key and value * 2 Threads were inserting and getting values from the cache Test results: * https://docs.google.com/spreadsheets/u/1/d/1ILk1gqdtY8bYArfwKCg5DBcjLO1gs... * For 700 MB Container, the calculated eviction size is 190 MB. * For 512 MB, it's ~59 MB. * 430 MB is the smallest container you can get with those values. Otherwise negative eviction parameter is calculated, which is not valid. * 1 GB container was ran using 10K keys and values. Calculated eviction size was 346 MB. [~william.burns] mentioned that: * Off-heap scales pretty good and the lowest observer overhead was 100 B/entry. * There is no noticeable difference between 1 and 2 threads in the test. * Someones eviction doesn't work for some reason. * Calling {{clear}} might be a potential problem because it load all entries on heap. Full explanation has been described [here|http://post-office.corp.redhat.com/archives/jboss-clustering-team/20...]. The calibration work will be continued in ISPN-8540. -- This message was sent by Atlassian JIRA (v7.5.0#75005)