RFR: Traversal: Purge arraycopy 'optimization'

[Aleksey Shipilev]

On 08/02/2018 10:54 PM, Roman Kennke wrote:
> We have an 'optimization' for arraycopy and TraversalGC: we attempt to
> offload the actual scanning+traversing of copied arrays to GC workers.
> However, there a a number of problems with this approach:
> - Termination: mutators can keep pushing work to GC workers, which would
> then not terminate and run into degen and/or full-gc. I've actualy seen
> this with some workloads.
> - Locking: arraycopy barrier code is called as leaf, which means it must
> not use locks. bad bad us.
> - More locking: this stuff is not scalable. we can currently only keep
> threads from trampling over each other feet by cranking up the threshold.
> 
> Also, since we have not yet seen an actual benefit of the approach,
> let's drop it. If we ever see arraycopies to be a problem that needs
> fixing, we can find it in the hg history.
> 
> http://cr.openjdk.java.net/~rkennke/traversal-kill-arraycopy/webrev.00/

Looks good, do it.

-Aleksey