RFR: Adaptive/Traversal heuristics rewrite for allocation rate

[Aleksey Shipilev]

On 08/22/2018 12:48 PM, Roman Kennke wrote:
> The patch looks good.
> 
> - Have you looked at it in visualizer? Does it look reasonable?

Yes, I did. Yes, it does. I also had specjbb runs before/after and GC times, counts, CSet sizes
appear the same. Which means both current and new heuristics converge to the same.

> - Have you actually tried with high-LDS workloads (ideally in visualizer)?

The key here is not high-LDS, but rather the erratic allocations when there is little room for
error. I ran a few specjvm tests on low heaps that do this, and the results were favorable for this
new heuristics.

I am planning to do more experiments and tuneups to this, as we go.

-Aleksey