In the GC logs for the ATG CRM demo we were seeing a fair number of "premature" evacuations (i.e. evacuation pauses that seemed to occur back-to-back with another evacuation pause with little or no application in-between and an eden occupancy that was much much less than capacity).

In a typical GC log around 30 such pauses were being seen.

Printing the GC cause indicated that all of these premature evacuations were being instigated by the GC locker code.

This issue was reported as 7143858. And turned out to be a race between an allocating thread requesting a GC and the GC Locker requesting GC.

Even after the 7143858 was fixed, however, we are still seeing GC logs from ATG that include some premature evacuations (3 vs. 30) and they are still GCs being instigated by the GC Locker.