Survivor rate groups are used by G1 to determine how many objects has been survived for a given region allocated at time X during mutator time.

I.e. we track copied objects for the first region allocated, for the second, third and so on, and track the amount of objects that had to be copied around, assuming that this value will stay around the same over time.

We do that both for eden and survivor regions.

While above assumption seems valid for eden regions as they are newly allocated each time, it is unlikely that the assumption holds for survivor regions - survivor regions contents are objects with random age as they are copied without regard to their original age to these regions. So a survivor region filled in gc X may have (and likely has) a completely different object composition (in terms of age) than a survivor region at gc X+1.

The age in the objects should be sufficient to determine object death rates within survivors.

Investigate removing region based survivor rate tracking for survivor regions, and find alternatives (i.e. age table) any use of that data for prediction.

Removing this tracking can be useful for simplifying code e.g. for JDK-8220089.