The JTREG tests for the SurvivorAlignmentInBytes option test success/failure on the number of bytes the objects take after copying them around. In addition to that there is some slack that accounts for objects potentially created by VM agents, i.e. extra bytes copied around. It reports these additional bytes copied as "someone else allocated objects".

The problem is that with the current amount of slack, most of the time the test succeeds, because it detects that "someone else copied a lot" (even without any agent).

The task is to somehow find an alternative way of detecting correct functioning and avoid false failures at the same time.