Deallocating memory pages
yamauchi at google.com
Thu Feb 21 23:42:04 UTC 2013
> This is really unfourtunate, from my POV especially for client-side workloads.
> On the desktop, footprint is often more important than
> peak-performance, especially when running many java based desktop
> (although I have to admit CMS doesn't make a lot of sence there).
I am inclined to agree with that. Re: CMS, I'm curious, does CMS not
work well for client-side apps?
> What I find so intriguing about this patch is that it basically
> reduces one of the drawbacks of GC based systems: Unused memory is not
> given back to the memory manager of the OS. I've recently read a paper
> about the idea of a GC interface provided by the kernel to avoid this
> issue, but this patch essentially achieves the same just by using
> existing interfaces.
BTW, would anyone like to try out this patch and give feedback? It's
at http://cr.openjdk.java.net/~hiroshi/webrevs/dhp/webrev.00 (which is
currently Linux/CMS only and needs to be patched into
http://hg.openjdk.java.net/hsx/hotspot-gc/hotspot/ and built).
More information about the hotspot-gc-dev