Submitted JEP 189: Shenandoah: An Ultra-Low-Pause-Time Garbarbage Collector

[Roman Kennke]

Hi,

> I read that Brooks pointers have not be often be considered because of 
> the high cost associated with indirection, but a blog from Roman 
> (https://rkennke.wordpress.com/2016/02/08/shenandoah-performance/) 
> seems to show that perhaps this belief should be re-evaluated. 
>
> Do you have any further insights that you can share on the good 
> performance of Shenandoah ? 

A straightforward implementation of the barriers that are necessary will indeed show disappointing performance. We've spent the better part of Shenandoah's development on optimizing the barriers in C2 and are now able to often hoist barriers out of hot loops or eliminate them altogether.

Cheers, Roman